Increasing freshwater and dissolved organic carbon flows to Northwest Alaska's Elson lagoon

This study focuses on terrestrial exports of freshwater and DOC to Elson Lagoon in northwest Alaska. The spatial domain encompasses the watershed that drains the small rivers and streams connecting to the lagoon. The 'Elson Watershed' is approximately 579 km² and is partitioned into 579 1 × 1 km grid cells (figure 1). A numerical model simulation was executed over the period 1981–2020 with the Permafrost Water Balance Model (v4) (figure S1 (available online at stacks.iop.org/ERL/16/105014/mmedia)). The transient 40-year simulation was preceded by a 50 year spinup on year 1980 to stabilize soil temperature, moisture, and soil DOC pools. Model parameterizations for soil texture, vegetation cover, and other static field are described in (Rawlins et al 2003, 2013). Vegetation across the watershed is exclusively tundra. While parameterizations for fields such as soil texture and vegetation cover are fundamental elements of land surface and hydrological model simulations, simulated runoff in Arctic regions is most sensitive to the time-varying meteorological forcings (Rawlins et al 2003). In the northern high latitudes a sparse station network often hinders confident assessment of the quality of gridded fields of near-surface (2 m) air temperature and precipitation necessary for model simulations across broad regions. This uncertainty is ameliorated in the present study through the use of meteorological observations at Utqia$\acute{\text g}$vik, Alaska, the most northerly first-order station (71.8189 ^∘N, 156.8479 ^∘W; 9.5 m ASL), in operation by the National Weather Service since 1901. Given the relatively flat terrain and small watershed size, daily 2 m air temperature, precipitation (available at http://climate.gi.alaska.edu/acis_data), and wind speed (available at https://mesonet.agron.iastate.edu) (table 1) were assigned to each of the watershed's grid cells.

Zoom In Zoom Out Reset image size

The PWBM was recently updated to add process representations related to water loss from the surface and DOC production, decomposition, storage and loading to river networks (Rawlins et al 2021). Briefly, DOC production is influenced strongly by the amount of surface and soil organic matter and the dynamics of surface/soil temperature and moisture. A production rate coefficient (k_prod , table S1) is adjusted during calibration. The decomposition of DOC is assumed to result in losses to carbon dioxide and/or sorption to the mineral soil. As with production, a decomposition rate coefficient (k_decomp ) is adjusted during calibration. Daily DOC production and loss to decomposition are used to update a soil DOC pool. Transfer of DOC from the surface or a soil layer to the stream and river network takes place whenever surface or subsurface runoff occurs:

$$\begin{equation} DOC(t) = S_{DOC}(t) \, Q, \end{equation} \tag{ 1 }$$

where DOC(t) is mass loaded to a river network (g C m² day⁻¹), $S_{DOC}(t)$ is soilwater DOC storage (g C m⁻³), and Q is runoff (m day⁻¹). Rawlins et al (2021) provide additional detail on the recent model updates.

Considerable uncertainty exists in the density of soil organic carbon in regional-scale simulations across Arctic environments. Given the relatively small extent of Elson Watershed, soil carbon density for each grid cell was assigned 50 kg m⁻³, which was the areally weighted mean amount calculated from soil cores collected along a transect within the watershed (Bockheim et al 1999), and is an approximate mid-point value between the 30 cm (24 kg m⁻³) and 100 cm layer (78 kg m⁻³) estimates for the area from the Northern Circumpolar Soil Carbon Database (Hugelius et al 2013).

The production, decomposition, and subsequent leaching of DOC into surface runoff is simulated in a similar manner to soil processes. In contrast to soil organic carbon, a surface carbon pool is parameterized as aboveground biomass from available data. This parameterization is important, as river sampling has consistently shown that DOC concentrations of North Slope and other arctic rivers peak during the spring freshet, when snowmelt-driven runoff leaches decaying tundra vegetation and organic-rich shallow soils above frozen ground, thereby mobilizing massive quantities of DOC that is shunted through fluvial networks (Spencer et al 2008, Holmes et al 2012, McClelland et al 2012). Aboveground carbon is parameterized here based on estimates of biomass from sites originally established near Utqia$\acute{\text g}$vik in 1972, and resampled in 1999, 2008 and 2012 (Lara et al 2012, Villarreal et al 2012). Biomass from the vegetation communities in these plots averages 49.9 g C m⁻² as reported by Goswami et al (2015). Parameterization of aboveground carbon storage is obtained by applying the rate increase derived from satellite-based estimates of net primary productivity (NPP) for the period 1985–2005 (available at http://files.ntsg.umt.edu/data/PA_Monthly_ET/) (table 1) (Zhang et al 2008) to the vegetation communities plots average set at its midpoint year 1992. Rawlins et al (2021) provide additional details of the model algorithms involving DOC production and leaching to streams and rivers.

The change analysis is focused on annual and seasonal riverine freshwater and DOC export totals. Recent evidence of hydrological cycle intensification and permafrost thaw across the region provide a basis for positing a null hypothesis of increasing freshwater and DOC flows to Elson Lagoon. Trend significance is assessed with the Mann–Kendall test (Mann 1945, Kendall 1955), with p-values below p = 0.05 considered to be statistically significant. Temporal autocorrelation for each time series examined was first assessed with the Durbin–Watson test (Durbin and Watson 1950), which involves the residuals from ordinary least squares regression:

$$\begin{equation} D = \frac{\sum_{t = 2}^{n} (e_{t} - e_{t-1})}{\sum_{t = 1}^{n} e_{t}^{2}}, \end{equation} \tag{ 2 }$$

where e_t = $y_t - \hat{y_t}$ are the residuals from the ordinary least squares fit of the y_t dependent variable, and $\hat{y_t}$ are the predicted values from the fitted linear regression model. The D test statistic values were compared with published tables of critical values to accept or reject the null hypothesis of no evidence of positive or negative autocorrelation at significance level α = 0.05. None of the time series examined were found to exhibit positive serial autocorrelation, though the test was inconclusive for some variables. For surface and total DOC export, D values slightly above the critical value for negative autocorrelation imply a slight underestimation of the level of statistical significance. Given these results no transformations were performed on the variables examined.

Measurements of river discharge from the small rivers and streams which drain to Elson Lagoon have only recently been made, and no long-term data for the area exist. In light of this shortcoming, the magnitude in simulated annual runoff is evaluated here against averages described in Arp et al (2020) for the Fish Creek Watershed, located approximately 200 km to the southeast. Simulated runoff averages of 91 (2001–2008), 129 (2009–2015), and 150 mm yr⁻¹ (2016–2019) compare favorably to the 90, 120, and 163 mm yr⁻¹ for Fish Creek, with percent differences of 1.1%, 7.5%, and 8%, respectively. This result suggests that the model simulation broadly captures the magnitude of freshwater export to Elson Lagoon. Similar model performance was demonstrated in a study of hydrological quantities estimated over the full North Slope drainage basin. For example, model simulated discharge for the Kuparuk River of 1.3 km³ yr⁻¹ (1981–2010) aligns with the observed annual total from USGS gauge data of 1.4 km³ yr⁻¹ (Rawlins et al 2019).

As with river discharge, there is a paucity of field observations of DOC concentration from streams and rivers that drain to the lagoon, particularly during the spring freshet, a shortcoming that hinders a robust validation of simulated DOC concentrations and associated DOC export. Here simulated DOC concentration for all rivers draining to the lagoon are compared to estimates for spring (May–June) and summer (July–September) that were derived from an empirical relationship between watershed slope and concentrations obtained from sampling of moderate and large Arctic rivers (Connolly et al 2018). The spring and summer periods were defined in Connolly et al (2018) to capture the spring (snowmelt-driven) freshet and subsequent summer low flows. Median simulated spring and summer DOC concentrations over years 2016–2020 of 24.3 and 8.2 mg C L⁻¹, respectively, are broadly consistent with the independent empirical averages of 18.0 and 13.2 mg C L⁻¹ for the region. While the empirical estimates were obtained from a wide range of river measurements throughout the North Slope and other parts of Arctic, rivers across Northwest Alaska were relatively underrepresented in the construction of the empirical functions. Simulated concentrations of 5–10 mg C L⁻¹ during late summer (late July to mid August) are also well within the range of 3–13 mg C L⁻¹ obtained from sampling of small rivers near the lagoon at that time (Lougheed et al 2020).

End of season maximum active layer thickness (ALT) derived from simulated soil temperatures are congruent with data from the Circumpolar Active Layer Monitoring (CALM) program (available at https://www2.gwu.edu/calm/data/north.htm.) (Hinkel and Nelson 2003). Simulated ALT averaged across the watershed generally falls between the CALM observations for Barrow and Atkasuk (figure S2). For 2017 simulated ALT is in the upper half of the distribution of 30 m ALT from Remotely Sensed Active Layer Thickness (ReSALT) data (available at https://doi.org/10.3334/ORNLDAAC/1796) (Schaefer et al 2015). Each data series shows increasing ALT over time. Simulated ALTs are also consistent with the range reported in field studies (Nelson et al 1997, Hinkel et al 2001, Jafarov et al 2017).

As with other parts of the Arctic, air temperatures in and around Utqia$\acute{\text g}$vik AK have warmed markedly over the past four decades. Since 1981 average temperatures have warmed significantly (p < 0.01) in all seasons (table 2, figure S3). Annual, winter, and spring average air temperatures have warmed at a rate of 1.1, 1.1, and 1.0 ^∘C decade⁻¹ (p < 0.01). Autumn has warmed by 2 ^∘C decade⁻¹, nearly twice the rate of winter warming. The autumn temperature average is −3.5 ^∘C over the most recent five year period 2016–2020. Given the long-term trend rate it is conceivable that a typical autumn will average above freezing by mid century. Summer also exhibits a significant positive trend, though considerably less (0.4 ^∘C decade⁻¹) than the three other seasons.

Precipitation increases are noted in all seasons (table 2, figure S4), with the changes statistically significant for annual, winter, spring, and autumn totals. The trend in summer precipitation (5.1 mm decade⁻¹) falls just outside the level of statistical significance, though precipitation over the most recent 5 year period (2016–2020) is more than 50% greater than the first 5 years (1981–1985), consistent with other studies pointing to increasing rainfall rates across parts of the high northern latitudes and river discharge from watersheds approximately 200 km to the east of Elson Lagoon (Arp et al 2020). Precipitation over the latter period is nearly double the early average in winter, spring, and autumn, with percent changes of 103%, 85%, and 85% respectively. Annual total rainfall more than doubled between the first and last five years and increased significantly (p < 0.001) at a rate of 10.8 mm decade⁻¹ over 1981–2020. Autumn shows the greatest rate of change for average air temperature (2.0 ^∘C decade⁻¹). The rate increase in seasonal total precipitation is also highest in autumn (8.2 mm decade⁻¹). This accelerated change during autumn is an expected manifestation of the marked sea ice losses that have occurred in the nearby Beaufort and Chukchi Seas (Frey et al 2014, Steele et al 2015).

Over the simulation period 1981–2020, increases in surface and subsurface runoff across the contributing watershed are evident (figure 2). The increases in runoff—17.7, 11.7, and 6.0 mm decade⁻¹ for total, surface, and subsurface runoff, respectively (table 2)—and associated freshwater export to Elson Lagoon are consistent with recent increases in discharge from rivers in the Fish Creek Watershed (Arp et al 2020). In particular, the excessive amount of subsurface runoff in 2019 (126 mm) aligns with the highest runoff reported for Fish Creek Watershed's rivers over the 2001–2019 period examined in that study. These trends point to hydrologic cycle intensification manifesting in increased freshwater export to Elson Lagoon and nearby parts of Northwest Alaska.

Zoom In Zoom Out Reset image size

Organic carbon stored in tundra regions is highly vulnerable to decomposition through warming-induced thaw of permafrost (Xue et al 2016, Feng et al 2020). Organic carbon in the active layer is also vulnerable to increased decomposition rates under warmer conditions. Soil temperatures near Elson Lagoon have warmed markedly over recent decades. For example, for a representative grid in the watershed, soil temperature at 13 cm depth between DOY 170 and 240 averaged 0 ^∘C in 1983 and 4 ^∘C in 2019 (figure 3). Soil temperature at 23 cm remained near 0 ^∘C during summer in 1983, a result of the 'zero curtain', the effect of latent heat maintaining temperatures near 0 ^∘C over extended periods. In 2019 soil temperature at 23 cm averaged 3 ^∘C, with thaw at the end of summer reaching close to 50 cm. Assuming that the active layer extends to the depth at which soil temperature equals 0 ^∘C, maximum ALT, as an average of all of the watershed's grids, averaged 28.5 cm over the simulation's first five years (1981–1985) and 45.8 cm over the last five years (2016–2020) (figure S2). The trend over the full 40 year period is statistically significant (p < 0.001) for all grids, averaging 4.0 cm decade⁻¹ (s.d. 0.51 decade⁻¹). Soil moisture in the thawed active layer reflects the influence of climate warming. For example, in the years 1981–1983, nearly saturated soil is present through the profile during the first few weeks after thaw and in the lower half of the profile in late summer (figure 4(a)). The active layer extends to a depth of 20 cm in 1981 and 1982, and only 10 cm in 1983. The simulation shows a much deeper depth of thaw in 2017–2019 (figure 4(b)). In the latter period thaw extends to roughly 40, 30, and 50 cm in 2017, 2018, and 2019 respectively. Soil water content increases from the surface downward, consistent with field measurements that captured water pooling above the permafrost (Hinzman and Kane 1991). This gradient expressed in the model simulation is expected, and noted in the earlier period as well, though not as pronounced given the shallower depth of thaw. Surface warming along with drainage into a deeper profile contribute to enhanced drying of near surface soils near the end of summer in more recent years. Warmer soils and a larger volume of thawed material manifest in a trend to greater production of DOC in the subsurface flow (figure 5(a)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The production of DOC in soils varies with SOC content, and is sensitive to soil temperature and moisture. Simulated production is low in winter when soils are cold and there is little or no liquid water present (figure S5(a)). During the warm season production varies by a factor of five, and is highest when the temperature of a model soil layer approaches 15 ^∘C, and it is nearly water saturated, averaging approximately 10 mg C day⁻¹ during these conditions. The simulation demonstrates how the magnitude of DOC leachate that contributes to stream loadings is lowest when the soil water DOC pool is at the low end of its range, and when the concentration of the leachate is low (figure S5(b)). Conversely, leachate magnitudes are highest when soil water DOC storage and leachate concentrations are high, exceeding at times the daily production rate, particularly when soils are nearly saturated during runoff events, for example, during or following heavy rains in the warmth of late summer (figure S6). Soil water DOC storage decreases sharply in late spring during flushing events to roughly half its magnitude prior to the freshet. Production of DOC in summer increases the storage pool, with the change in magnitude reflecting both production and decomposition. The storage pool also increases during winter, though influenced then by much lower rates of production and decomposition, relative to summer.

Over the past 40 years, DOC production at the surface and in soils increased (p < 0.01), with subsurface production exhibiting a slightly higher level of significance given lower interannual variability relative to the surface (figure 5). Simulated total DOC export to the lagoon likewise increased (p < 0.01), driven by increases from loading in surface (p < 0.01) and suprapermafrost flow (p = 0.01). The rate of change is slightly greater for surface export (table 2), owing to the influences of warming (figure S3(c)) and increasing biomass and surface runoff (figure 2). Significantly higher amounts of DOC flux in the suprapermafrost subsurface flow are noted in the more recent five year period relative to the past (figure 6). The change is greatest in late summer-fall. DOC export from suprapermafrost flow has increased in all months, except for June. Some 119.5 and 536.7 Mg DOC was exported in September and October during 1981–1985 and 2016–2020, respectively, an increase of 450%. Daily mean export from suprapermafrost flow is now higher from June through September. Percent differences between the two periods range from 7.5% for July to 49.5% in September. There are now more days in May with DOC export from suprapermafrost flow (two in 1981–1985 vs. ten in 2016–2020), and many more days in October (three vs. ninety-eight). Suprapermafrost flow is now emerging in November. The changes in subsurface processes arise as a result of warmer soils, a greater volume of thawed soils, and increasing subsurface runoff.

Zoom In Zoom Out Reset image size

Approximately 34 km of coastline border Elson Lagoon. Subsurface flow along the coastal zone during summer provides a direct transfer of freshwater and DOC to the lagoon through the thawed active layer. The simulation indicates that late summer (August–October) suprapermafrost coastline freshwater export directly to the lagoon averages 6.4×10⁵ m³ during 1981–1985 and 1.7 × 10⁶ m³ during 2016–2020 (table 2), an increase of ∼165%. A massive flux anomaly of 3.3 × 10⁶ m³ in 2019 is a factor of four greater than the early 1980s average. Suprapermafrost coastal DOC export directly to the lagoon during late summer 1981–1985 varied from 3950–10 040 kg DOC, averaging 6180 kg C. In 2016–2020 this increased to 8280–23 840 kg C, with the highest amount exported in 2019. The 5 year average is 13 030 kg C, more than double (111%) the late summer export amount during the early period. The late summer total in 2019 represents an increase of 285% from 1981 to 1985 period average. September experienced the largest increases, with coastal suprapermafrost DOC export in 2019 (8906 kg C) a more than fourfold increase (459%) over the average during 1981–1985.