Climate, snowmelt dynamics and atmospheric deposition interact to control dissolved organic carbon export from a northern forest stream over 26 years

The Calumet watershed is a first-order tributary of Lake Superior in the Upper Peninsula of Michigan, USA (Latitude: 47.278, Longitude: −88.515), previously described in detail by Stottlemyer and Toczydlowski (2006). Briefly, the watershed is 176 ha, has uniform slope (5%), moderate topographic relief and is vegetated by mixed northern hardwood forest. Soils are Typic Haplorthods with an ortstein layer at 1.5–2.0 m depth (Stottlemyer and Toczydlowski 1991). The climate is modified by Lake Superior, which moderates autumn and winter temperature extremes (Stottlemyer and Toczydlowski 2006). The National Atmospheric Deposition Program (NADP) site (MI99) is located 16 km south of the watershed, and received a mean of 803 mm in annual precipitation over the study period (supplemental figure; figure S1 (available online at stacks.iop.org/ERL/15/104034/mmedia)). The annual hydrograph is dominated by snowmelt and can be highly variable in timing and duration. The transitional spring climate has a large impact on the timing and duration of the spring snowmelt hydrograph (figure 1).

Zoom In Zoom Out Reset image size

Snow cores were measured weekly typically from mid-December to mid-April across four transects spanning the elevation gradient at Calumet, 1988–2013 (Stottlemyer and Toczydlowski 1991). The transects consisted of five sampling points and were measured weekly using a Federal sampler for snow moisture content (Stottlemyer and Toczydlowski 2006). These transects were averaged for a watershed SWE estimate. Continuous stream discharge was monitored using a 30 cm wide Parshall flume equipped with a Li-Cor datalogger and Stevens pressure transducer. Continuous data were integrated by day to calculate a daily measurement of runoff. Stream water samples were collected approximately weekly except during snowmelt, when samples were collected multiple times per week to capture the leading and trailing edges of the hydrograph. Water samples were collected in amber polyethylene 500 ml bottles and frozen for later analysis. Precipitation chemical composition data was acquired from the NADP monitoring site MI99 in Houghton, MI and was used to obtain monthly averages of ${\text{SO}}_4^{2 - }$ deposition (wet deposition) for the study period (1988–2013). Daily temperature data was collected from the National Oceanic and Atmospheric Administration (NOAA) station 3908 Houghton-Hancock County Airport 8 km south of the watershed.

Water samples were immediately brought to Michigan Technological University for analysis. DOC sample analysis is described in detail by Stottlemyer and Toczydlowski (2006). Briefly, for the period (1988–1997) DOC was determined on filtered samples (0.45 μm) using a Dohrmann 180 (Teledyne-Tekmar-Dohrmann, Mason, Ohio) carbon analyzer. Post 1997, DOC concentrations were determined using a Shimadzu TOC-5000A analyzer (Shimadzu Scientific Instruments, Columbia, Maryland). Laboratory quality assurance procedures were previously described in detail (Stottlemyer and Toczydlowski 2006), and included analyses of split samples run as part of NADP, the Environmental Protection Agency's National Acid Precipitation Assessment Program (NAPAP) and continuous participation in the U.S. Geological Survey laboratory round robin program.

2.4.1. Snowmelt descriptors

2.4.2. Statistical analysis

Variation in the onset of melt, and hence the duration of winter, as well as snowpack depth have changed over time. The occurrences of winters associated with high maximum SWE and extended snowmelt through the spring have declined over the study period (Bartlett test; p = 0.023; n = 26 years) (figure 2(a)). Annual cumulative runoff exhibited a marginally significant downward trend of 2.4 mm Y⁻¹ (Standard Error, SE = 1.2 mm Y⁻¹) (R² = 0.11, p = 0.050; figure 2(b)). Steep time also showed decreasing variation through time, indicating spring climate is becoming more homogenous with less frequent extended snowmelt periods (Bartlett test; p = 0.033; n = 26 years) (figure 2(c)). There were no trends in temperature within the range of records reflected in this study (1988–2013). Temperature records extend beyond the solute records in this study (1948–2018), and there has been a long-term increase in air temperatures observed in August and September of 0.026 °C Y⁻¹ (SE = 0.010 °C Y⁻¹) and 0.034 °C Y⁻¹ (SE = 0.009 °C Y⁻¹), respectively (figure S2). However, annual mean temperature did not show a significant warming trend (mean of 4.5 °C).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Stream DOC concentration ranged from 2.61 mg C l⁻¹ to 20.00 mg C l⁻¹ (mean = 8.12 mg C l⁻¹ over the course of the study). Mean annual DOC concentrations increased significantly by 0.14 mg C l⁻¹ Y⁻¹ (SE = 0.03 mg C l⁻¹ Y⁻¹) (figure 2(d)). ${\text{SO}}_4^{2 - }$ concentrations in precipitation have been decreasing by approximately 0.026 mg l⁻¹Y⁻¹ (SE = 0.003 mg L⁻¹ Y⁻¹) in Northern Michigan (1988–2013; National Atmospheric Deposition Program; MI99; figure S3). Path analysis indicated that ${\text{SO}}_4^{2 - }$ concentration had a direct negative effect (−0.57, p < 0.001) on DOC concentrations in the stream (figure 3). The path analysis also highlighted the negative effect of mean temperature on DOC concentrations (−0.58, p = 0.016), owing to the strong direct negative effects of mean temperature on maximum SWE (−0.71, p < 0.001) and in turn of maximum SWE on DOC concentrations (−0.77, p = 0.001).

Annual runoff showed a significant correlation with maximum SWE (R² = 0.57, p < 0.001; figure 4(a)). Annual runoff also correlated with annual DOC export (R² = 0.51, p < 0.001; figure 4(b)). There was a correlation between maximum SWE and annual DOC export (R² = 0.18, p = 0.018) and maximum SWE had a positive direct effect on DOC export in path analysis (figure 3). Because maximum SWE was strongly negatively related to mean temperature, temperature also had a negative indirect effect on DOC export in path analysis (−0.41, p = 0.002; indirect path data not shown). Melt onset date and peak discharge date correlated strongly with annual DOC export (figure 5), and as such later maximum discharge corresponded to higher DOC exports (significant positive direct effect, 0.24, p = 0.038; figure 3). In addition, maximum SWE was also positively related to the timing of maximum discharge in path analysis (0.82, p < 0.001). Steep time, or the duration of melt, was not significantly related to annual mean stream DOC concentration or export (p = 0.338 and p = 0.451, respectively).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

While snowmelt timing did not appear to directly influence annual stream DOC concentrations, factors related to snowmelt timing did correlate with DOC concentrations. Path analysis indicated that maximum SWE had a negative influence on DOC concentrations, likely through dilution effects (figure 3). However, the strength of this effect would be diminished if trends in maximum SWE continue to decline and if warming trends extend into the spring season. Mean temperatures had a negative effect on DOC concentrations in the path analysis conducted in this study. Prior research has observed an increase in DOC release in soils that experience freezing conditions, suggesting that the integrated effect of air temperature and winter snow cover duration drives soil frost formation and subsequently stream DOC concentrations (Haei et al 2010). The mechanism by which soil freezing increases DOC concentrations is likely a combination of increased solutes from root and microbial mortality (cell lysis) and solute concentration by freezing (Biederbeck and Campbell 1971, Fitzhugh et al 2001, Agren et al 2012). In addition to these mechanisms, we highlight here that soils in this study region typically do not freeze in the winter because of the depth and persistence of the snowpack; this effectively disconnects winter soil and air temperatures. It is possible that warmer fall temperatures (cf figure S2) could delay snowpack formation, which counterintuitively could permit freezing conditions in soils (Groffman et al 2001) and thus promote the release of DOC (Fitzhugh et al 2001, Haei et al 2012).

Stream DOC concentrations have been steadily increasing over the last three decades, while atmospheric ${\text{SO}}_4^{2 - }$ concentrations have declined. The steady decline in ${\text{SO}}_4^{2 - }$ deposition offers a parsimonious explanation for the increase in observed DOC concentrations (Ledesma et al 2016). In addition, we highlight here that runoff appears to affect DOC concentrations and water delivery that drives export. Future work is needed to tease apart the complex role climate plays in mediating runoff, particularly in snowmelt dominated watersheds. These findings are in line with previous studies that have identified atmospheric deposition as the principal driver of increased DOC concentrations in surface waters, while highlighting the important influences of climate factors with strong seasonal effects on carbon concentrations (see Clark et al 2010, and references therein). For example, seasonal changes in snowmelt flushing and precipitation events create considerable variation in DOC concentrations (cf coefficients of variation ranging from 27%–39%; Creed et al 2003), with important implications for DOC export dynamics.

Winter snowpack has the largest direct effect on DOC export through runoff (figure 4). Additionally, snowmelt timing shows a positive correlation with annual DOC export for the watershed, which is also influenced by snow water equivalences (figure 3). These results indicate that the amount of water in the snow is more significant than the timing of snowmelt in controlling DOC export. However, temperature appears to have a greater effect on timing than snow water equivalence (figure 3), which could alter the relative magnitudes of these effects in the future. The decline in runoff over time (figure 2(b)) along with the direct and indirect effects of snowmelt and temperature on export (figure 3) offer countervailing effects which result in no time trend for DOC export. Other studies of temperate zone streams have also identified an increase in stream DOC concentrations with no associated increase in export (Urban et al 2011, Coble et al 2018, O'Driscoll et al 2018). Our analysis suggests that slight changes in runoff combined with indirect controls of temperature and snowmelt can offset effects of increased DOC concentrations and attenuate DOC exports. However, this is further complicated by the co-occurrence of slow-melt and low snowpack years (denoted by subscript letters (b) and (d) in figure 1; table 1), which add considerable variation to interannual DOC exports (figure 5).

The large proportion of seasonal precipitation falling as snow is a distinctive characteristic of watersheds in the northern Great Lakes region like the Calumet watershed, as it consistently provides a baseline minimum amount of water in spring runoff capable of flushing a minimum amount of DOC each year. While variation in snowpack among years was quite high in the Calumet watershed, the record shows there was always at least approximately 180 mm of water in peak snowpack (figure 2(a)). As such, runoff contributions from rainfall would have to be very large events—around 180 mm of precipitation in a given storm—to significantly alter the shape of the snowmelt-dominated hydrograph (e.g. figure 1). While larger and more variable storms are expected to occur more frequently in the Great Lakes region (Kossin et al 2017), the spring freshet is likely to remain the largest pulse of DOC export from watersheds in this region. Late fall runoff was generally the second largest pulse of DOC from the watershed. Late summer baseflow has been related to maximum SWE in mountainous regions (Godsey et al 2014, Barnhart et al 2016). In this study, July and August runoff showed significant linear trends with both maximum SWE (R² = 0.65; p < 0.001 and R² = 0.38; p < 0.001 respectively; figures S4 and S5) and melt initiation (R² = 0.64; p < 0.001 and R² = 0.37; p < 0.001 respectively; figures S4 and S5). Although speculative, this suggests that snowmelt recharge is significant for late summer baseflow conditions and subsequent DOC export. This highlights the importance of winter snowpack in these regions on the annual hydrograph and subsequent DOC exports.

Given climate change predictions for areas with high snowfall (e.g. Musselman et al 2017), we expect changes in the hydrographs of snowmelt-dominated tributaries, with consequences for solute export (cf Marcarelli et al 2019). Most notably, these predictions indicate earlier and slower snowmelt to prolong the hydrograph with an earlier melt onset. Although this 26 year study period contains more years in the late- than early-season melt scenario (15 and 11 years, respectively; table 1), our results indicate that both earlier and slower snowmelt may lead to decreased DOC exports. Exactly how climate-induced changes in the spring hydrograph affect DOC exports will depend on the persistence of increasing DOC concentrations, snowmelt dynamics and temperature. While trends to date are best represented as linear processes, we stress here that the increasing DOC concentrations and declining peak SWE may exhibit non-linear or threshold responses in the future. For example, runoff cannot continue to decline at the same rate indefinitely, and the increasing DOC concentrations should reach a new stable state over time (Monteith et al 2007b). As such, long-term monitoring efforts are invaluable for detecting novel patterns of DOC export.