Drivers of increased organic carbon concentrations in stream water following forest disturbance: Separating effects of changes in flow pathways and soil warming

Authors

  • J. Schelker,

    Corresponding author
    1. Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, Umeå, Sweden
    2. Now at Department of Limnology and Bio-Oceanography, University of Vienna, Vienna, Austria
    • Corresponding author: J. Schelker, Department Forest Ecology and Management, Swedish University of Agricultural Sciences (SLU), SE-901 83 Umeå, Sweden. (jakob.schelker@slu.se)

    Search for more papers by this author
  • T. Grabs,

    1. Department of Earth Sciences, Uppsala University, Uppsala, Sweden
    Search for more papers by this author
  • K. Bishop,

    1. Department of Earth Sciences, Uppsala University, Uppsala, Sweden
    2. Department of Aquatic Sciences and Assessment, Swedish University of Agricultural Sciences, Uppsala, Sweden
    Search for more papers by this author
  • H. Laudon

    1. Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, Umeå, Sweden
    Search for more papers by this author

Abstract

[1] Forest disturbance such as clear-cutting has been identified as an important factor for increasing dissolved organic carbon (DOC) concentrations in boreal streams. We used a long-term data set of soil temperature, soil moisture, shallow groundwater (GW) levels, and stream DOC concentrations from three boreal first-order streams to investigate mechanisms causing these increases. Clear-cutting was found to alter soil conditions with warmer and wetter soils during summer. The application of a riparian flow concentration integration model (RIM) explained a major part of variation in stream [DOC] arising from changing flow pathways in riparian soils during the pretreatment period (r2 = 0.4–0.7), but less well after the harvest. Model residuals were sensitive to changes in soil temperature. The linear regression models for the temperature dependence of [DOC] in soils were not different in the disturbed and undisturbed catchments, whereas a nonlinear response to soil moisture was found. Overall these results suggest that the increased DOC mobilization after forest disturbance is caused by (i) increased GW levels leading to increased water fluxes in shallow flow path in riparian soils and (ii) increased soil temperature increasing the DOC availability in soils during summer. These relationships indicate that the mechanisms of DOC mobilization after forest disturbance are not different to those of undisturbed catchments, but that catchment soils respond to the higher hydro-climatic variation observed after clear-cutting. This highlights the sensitivity of boreal streams to changes in the energy and water balance, which may be altered as a result of both land management and climate change.

1 Introduction

[2] There has been a recent increase in research efforts to understand and quantify the role of streams and rivers in the global carbon (C) cycle [Cole et al., 2007; Lauerwald et al., 2012; Wallin et al., 2013]. Within this effort, boreal catchments are especially important, because of their large spatial extent and generally high concentrations of dissolved organic carbon (DOC) [Laudon et al., 2012]. In addition, they are under sustained pressure from land management that can fundamentally impact the mobilization of DOC from terrestrial ecosystems into aquatic environments [Bolan et al., 2011; Mattsson et al., 2005; Stanley et al., 2012].

[3] The regulation of stream DOC concentrations in northern catchments, even in areas that have not been subject to large-scale anthropogenic disturbance, are complex. However, two main characteristics determine the range of DOC concentrations in streams. First, the size and location of soil pools of organic matter (OM) [Aitkenhead et al., 1999], which include carbon rich A-horizons that dominate upland contributions [Raymond and Saiers, 2010], peat rich riparian soils [Creed et al., 2003; Grabs et al., 2012] as well as peatlands and mires [Dinsmore et al., 2010; Fenner et al., 2004] are of importance. Second, the transport of OM from “DOC source areas” [Burns et al., 2013; Schiff et al., 1998] to streams is dependent on a number of biogeophysical processes. First, DOC mobility may depend on chemical mobilization and immobilization in the soil solution [Kalbitz et al., 2000] which may vary as an effect of changes of catchment acidity [Evans et al., 2006; Kerr and Eimers, 2012; Löfgren et al., 2010; Monteith et al., 2007]. Further, DOC mobility in soils may be controlled by the variation in hydrological flow pathways of hillslopes [Weiler and McDonnell, 2006], the hydrologic connectivity of different landscape units [Laudon et al., 2004; McGlynn and McDonnell, 2003; Schelker et al., 2011], as well as the dynamics of shallow groundwater levels [Bishop et al., 2004; Grabs et al., 2012; Seibert et al., 2009; Xu et al., 2012]. In addition, seasonal controls are also important factors, such as, the variation in soil temperatures may promote higher DOC availability in soils [Freeman et al., 2001; Köhler et al., 2009; Winterdahl et al., 2011a] or increased soil frost can increase the soil and stream DOC levels during the spring freshet [Ågren et al., 2012; Haei et al., 2010].

[4] The question of what drives commonly observed increases in DOC concentrations after ecosystem disturbance, such as forest harvesting, is subject to recent investigations [Bolan et al., 2011; Schelker et al., 2012; Stanley et al., 2012]. Forestry operations are followed by a period of lower nutrient and water uptake by the remaining vegetation [Grant et al., 2007; Palviainen et al., 2005]. The following 5–20 year period of forest reestablishment leads to increased leaching of various nutrients into surface waters [Kreutzweiser et al., 2008; Löfgren et al., 2009]. In addition, a number of biophysical variables change the local conditions of the catchment soils: First, soil temperatures and soil moisture in clear-cuts are higher during summer [Olchev et al., 2009; Schelker et al., 2012], largely dependent on the higher soil irradiance and the lower transpiration in harvested areas. Second, the water cycling in soils is altered. During spring snowmelt higher snow accumulation [Boon, 2012; Jost et al., 2007; Varhola et al., 2010] and differing melting patterns in clear-cuts may increase stream runoff at least during some years [Penn et al., 2012; Schelker et al., 2013]. However, the largest impact of forest clear-cuts on soil and stream hydrology is observed during summer, where increased discharge has been quantified [Andréassian, 2004; Bosch and Hewlett, 1982; Hornbeck et al., 1997]. These increases are present as higher event runoff as well as higher GW levels [Monteith et al., 2006] that in turn can result in enhanced DOC flushing into streams during events [Boyer et al., 1996; Grabs et al., 2012; Laudon et al., 2011; Schelker et al., 2012; Xu et al., 2012]. Third, logging residues may represent an additional soil C-pool [Hazlett et al., 2007; Hyvonen et al., 2000] that can potentially contribute to increased DOC concentrations in catchment soils as well as in adjacent streams.

[5] Although it is not clear what the most important mechanisms of the increased mobilization of DOC after clear-cutting are, forest clear-cutting in the boreal region has been observed to result in increased DOC concentrations for at least a decade after harvesting [Kortelainen and Saukkonen, 1998; Kreutzweiser et al., 2008; Lamontagne et al., 2000; Laudon et al., 2009; Nieminen, 2004; Piirainen et al., 2007; Schelker et al., 2012]. As a consequence of the large geographical scale of forestry in many boreal areas, forestry can be considered as one of the largest potential anthropogenic disturbances of streams in high latitude regions.

[6] In this study, we are specifically addressing the question of what drivers are responsible for increases in [DOC] in boreal first-order streams that are subject to particular forestry operations (clear-cutting and the following site preparation before replanting). We hypothesize that the regulating mechanisms of DOC remain the same as for undisturbed boreal catchments, but that the increased GW levels and more shallow flows in riparian soils increase the mobilization of DOC into streams after forestry operations. Based on previous observations, we hypothesize that (i) water pathways through catchment soils are a first-order control for the increased DOC mobilization to streams and (ii) that soil temperature and soil moisture status represent a second-order control for the impact of forestry operations on stream DOC levels during the first decade after harvesting.

2 Methods

2.1 Study Site

[7] This study was conducted in the “277 Balsjö paired catchment experiment” located in the boreal forest of northern Sweden (N 64° 1′ 37″ E 18° 55′ 43″; Figure 1). The experiment, which was established in 2004, is situated approximately 70 km from the Baltic Sea coast. The main aim of the experiment is to quantify the effects of forestry operations on the hydrology and biogeochemistry of a boreal forest ecosystem [Löfgren et al., 2009; Schelker, 2013].

Figure 1.

The Balsjö paired catchment experiment, northern Sweden [Löfgren et al., 2009]. Thin black lines represent the stream network, and thick black lines represent the catchment boundaries; filled black dots indicate the location of sample drawing and stream gauging stations for the CC-4, NO-5, and NR-7 catchments, respectively. The NR-7 catchment is a nested subcatchment of NO-5. Wetlands are shown as light grey, clear-cut harvests as dark grey. Grey triangles mark the locations of the soil temperature and soil moisture measurements within the NR-7 control forest (TDR-NR) and the NO-5 clear-cut area (TDR-CC). GW-transects with continuous GW level measurements during spring to fall 2009 are shown as dotted white circles.

[8] The geographical setting of the Balsjö paired catchment experiment is typical for boreal forests of Fennoscandia. The vegetation (before the experimental treatments) was dominated by Scots pine (Pinus sylvestris) on well-drained upland locations, typically underlain by podsolic till soils (orthic podsols). In middle and lower elevations, Norway spruce (Picea abis) represents the most abundant tree species, whereas the near-stream zones were characterized by birch (Betula sp.). Riparian soils are peat dominated and peat layers can be more than 2 m thick (Lars Högbom, unpublished data). The glacially formed, low relief landscape is underlain by granitic bedrock (pegmatite, aplatic granite, and aplatite). Above the bedrock, highly compacted glacial till with low hydraulic conductivities forms the boundary between shallow and deeper groundwater.

[9] The experimental treatments of the three first-order streams that were part of this study include experimental clear-cut harvesting that covered 64%, 35%, and less than 5% of the of the catchment area of the CC-4, NO-5, and NR-7 catchment, respectively (Figure 1 and Table 1). Harvests were performed during early spring in 2006 on frozen ground with minor soil disturbance [Löfgren et al., 2009]. Clear-cutting was followed by common measures to promote forest regeneration performed in spring 2008. These measures, referred to as “site preparation,” included plowing (disk trenching) of the organic rich top soil to a depth of 5–10 cm on hillslopes. Wetter areas were left intact, whereas plowing in the remaining areas was performed in parallel to the hillsides to avoid downslope erosion. After that, Scots pine seeds were sown in furrows to promote forest regrowth [Löfgren et al., 2009; Schelker et al., 2012].

Table 1. Catchment Characteristics and Model Parameters
Catchment Characteristics:UnitCC-4NO-5NR-7
Catchment area(ha)40.539.824.2
Area harvested% area64%35%<5.0%
Number of observation wells for shallow GW-1164
Stream length affected by harvestm12505200
Riparian buffer-No, harvested to streamDiscontinuous; 10 m wide-
Model ParameterDescriptionUnit   
a (±SD)Change in flow at soil surfaces−1 m−1934 (±1075)1832 (±1614)16,584 (±32,946)
b (±SD)Slope of Q-GW relationshipm−113 (±5)17 (±3)14 (±7)
CdConcentration at depth dmg L−15.05.05.0
dBase level soil depthm−1.0−1.0−1.0
fShape parameter, best fitm−12.12.02.3

[10] Within the Balsjö paired catchment experiment, increases in [DOC] as a response to forest harvesting have been well described [Laudon et al., 2009; Schelker et al., 2012]. In short, Laudon et al. [2009] found significant increases of DOC concentrations of 23% in the two harvested areas compared to two control sites during the summer of 2007, one and a half-years after harvesting with a decreasing treatment effect during the fall. Furthermore, Schelker et al. [2012] showed than DOC concentrations for the three catchments used in this study averaged to 18.3 mg L−1 (n = 114), 18.2 mg L−1 (n = 112), and 18.6 mg L−1 (n = 114) for the CC-4, NO-5, and NR-7 catchment, respectively in the two year period prior to the harvest (2004–2006). After clear-cutting, but before site preparation, the concentrations increased to 21.6 mg L−1 (n = 215), 21.4 mg L−1 (n = 209), and 21.5 mg L−1 (n = 210) for the CC-4, NO-5, and NR-7, respectively. After site preparation, the DOC concentrations showed an additional increase at the treated sites. Average concentrations were measured as 31.5 mg L−1 (n = 265), 27.0 mg L−1 (n = 259), and 25.8 mg L−1 (n = 247) for CC-4, NO-5, and NR-7, respectively, during 2008–2009. These changes in average concentrations were primarily driven by responses to hydrological episodes. Average differences between the two clear-cut sites and the two control sites were between 8 and 15 mg L−1 higher during the spring freshet, whereas larger differences, between 5 and 24 mg L−1, were observed during rain episodes taking place during the summer and early fall [Schelker et al., 2012], indicating that seasonality may be an important driver to DOC variations after forest disturbance.

[11] Most productive forest land in Fennoscandia has been subject to human land management for at least the past one and a half centuries [Päivänen and Hånell, 2012]. Common practices included ditching and lowering of GW tables by deepening of streambeds in first-order streams. These measures were performed to increase forest productivity by increased drainage. The area of first-time drainage of Swedish forest land peaked in the mid-1930s when up to 500 km2 per year were ditched [Päivänen and Hånell, 2012]. Indications that similar action may have also taken place in the Balsjö catchment can be found by the fact that long sections of the streams are nearly straight at the same time as the low relief would suggest a more meandering shape of the watercourses (Figure 1). However, given that this land management by “shovel and spade” has not been documented in Balsjö in any form, precise estimates of the impact of this previous management on today's stream ecosystems remain difficult.

2.2 Data Collection

[12] This analysis was based on a total of 1579 stream DOC samples. Stream water sampling was performed as grab sampling between 2004 and 2011, with 1 to 2 week intervals during spring, summer, and fall, respectively, as well as 4 weeks between sampling occasions during winter low flow. Samples were stored cool and frozen within 2–3 days. The samples were then analyzed on a Shimadzu TOC 5050 analyzer before 2006 and on a Shimadzu TOC-VCPH analyzer after 2006. In addition, high-frequency samples of DOC were collected using ISCO-automated sampler units that were installed before snowmelt. ISCO units were then used continuously from the spring snowmelt until late fall (mid-October) of the years 2007–2009 with sampling frequencies of one sample every two days during low-flow conditions and up to three samples per day during runoff episodes. Similar to earlier work at this study site [Laudon et al., 2009; Schelker et al., 2012], organic carbon analysis was considered to represent DOC concentrations, even if samples were not filtered prior to analysis (see Schelker et al. [2012] for more details on sample collection and lab analysis).

[13] Ground water levels (z) were measured at a total of 23 locations within the three catchments during spring to fall of 2009. Measurements were performed using Trutrack (Trutrack Inc., New Zealand) water level loggers located in PVC-pipes that followed transects from the upland-transition zone (50–60 m distance from stream) to the stream (see also Figure 1). GW levels were recorded at an hourly basis and each GW time series was validated against manual GW water level measurements. Due to the fact that GW level dynamics at many locations were found to be correlated, a subset of six GW data sets was selected for the multivariate statistical modeling. For this selection, one GW logger representing the lower edge of the hillslope/riparian wetland transition zone (distance = 25–35 m from stream) and a second logger located within 0.5 m of the stream, representing GW levels in the stream near riparian zone in each of the three catchments, were selected.

[14] Soil temperature and soil moisture were measured with an approximate distance of 40–50 m from the stream, within the NR-7 (TDR-NR) and the NO-5 (TDR-CC) catchments (Figure 1). An automated station recorded hourly variations of soil temperatures (Ts) at 15 cm, 30 cm, and 50 cm soil depth, as well as variations of volumetric soil moisture (θ) as quantified by a series of Campbell-Scientific time-domain reflectometry (TDR) probes (model CS 616) at the soil depth of 20 cm and 50 cm. For each depth, data from three to four TDR probes were collected. Data used in the study represent averages of the available probes at the given depth. Probes with strongly deviating measurements were removed. Even if TDR probes were not locally calibrated, we consider the measurements to give a robust measure of changes in moisture dynamics between sites. Differences in soil temperatures (∆Ts), soil moisture content (∆θ), discharge (∆Q), and DOC concentrations (∆DOC) refer to values measured at the clear-cut site (within the clear-cut area of NO-5 for Ts and θ; at the outlet of CC-4 for [DOC] and Q) minus the value measured at the reference catchment (NR-7) at the same time.

2.3 Statistical Analysis and Calculations

[15] A subset of the data set for spring, summer and fall of the year 2009 was selected for statistical analysis. During this time period high resolution (hourly) GW levels (z), soil temperature (Ts), and soil moisture data (θ), respectively, stream discharge (log transformed; notation: logQ) as well as automated high frequency (48 h and 8 h) stream DOC sampling, respectively, were available (see section 2.2). A principle component analysis (PCA) for the 2009 data set (standardized; z-score) was performed to identify the relative importance of different drivers on the dynamics of DOC concentrations in stream water. This method uses the orthogonal variation in a data set to estimate how closely the different variables correlate in a linear, orthogonal model. Therefore, the PCA is insensitive to intercorrelation of variables (such as for example intercorrelated GW levels from different locations within the catchment). Further, the PCA allows setting the number of principle components, which is dependent on the dimensionality of the data set. After initial tests, a minimum of two output components was considered necessary for all catchments to reach correlation coefficients above the ones of the best simple linear regression models using only one predictor (common r2 = approx. 0.6 to 0.7 for simple linear regression models; see also section 3.2.). Simple linear regressions, Pearson's correlation coefficients, and Spearman rank correlation coefficients were considered significant if p < 0.05 and highly significant if p < 0.001. All statistical analyses and calculations were performed using MATLAB 7.8 (Mathworks Inc, Natick, MA, USA).

[16] Further, we defined the winter/spring season of each year as the time period from 1 January to 31 May. The summer/fall season was then defined as the remaining part of the year (1 June to 31 December). The entire study period included the years 2004–2011. The period during which soil moisture and soil temperature measurements were available included 2009 to mid-2011. Furthermore, we defined a first-order control as a factor explaining more than 50% of the variation of the response variable. Higher-order controls were defined similarly as a factor explaining more than 50% of the remaining variation, after accounting for all lower-order controls.

2.4 Static Model of Dissolved Organic Carbon in Riparian Soils

[17] In Swedish boreal catchments flow pathways in riparian soils change dependent on the hydrological conditions [Bishop, 1991]. Runoff generation in more shallow soil layers takes place during high-flow conditions such as the spring freshet or summer episodes when shallow GW levels rise and flush more surficial soil compartments. Given that DOC concentrations in riparian GW are highest at the soil surface and strongly decreasing with soil depth [Bishop et al., 2004; Grabs et al., 2012; Schelker et al., 2012], there is a direct hydrological control on stream DOC concentrations when flow pathways vary vertically throughout the year. To remove this “hydrological effect,” we applied the Riparian Flow-Concentration Integration Model (RIM) [Seibert et al., 2009] which allows to simulate the changes in stream DOC concentrations that are governed by the changes in shallow GW levels in riparian soils. Model residuals (measured minus modeled DOC concentrations) can then be used to investigate additional, higher-order drivers of DOC mobilization that would otherwise be masked by the variation in hydrological conditions. Earlier studies using this approach have been successful in highlighting the role of soil frost as a second-order control for spring DOC levels [Ågren et al., 2010], as well as the importance of soil temperatures for increased stream DOC concentrations during summer [Köhler et al., 2009; Winterdahl et al., 2011a].

[18] RIM is based on the assumption that the runoff present in the stream (Q) is equal to the integral of lateral inflows (q) from the riparian soils (equation ((1))) [Grabs et al., 2012].

display math(1)

[19] Assuming an exponential decrease of lateral transmissivity with soil depth (z), a common assumption for soils in northern Sweden [Seibert et al., 2003], the water flows may be expressed as q(z) = aebz, with a and b being model parameters describing the shape of the transmissivity profile. Considering an exponential decrease of [DOC] with soil depth, that is a common feature in northern Sweden [Bishop et al., 2004] and was also found in the Balsjö catchments [Schelker et al., 2012], the shape of the DOC profile in the riparian soils may be represented by csoilwater(z) = coefz, with c0 being the concentration at the soil surface and f being the shape parameter. The concentration in the stream may then be related to the soil solution concentration profile at any given time by equation ((2)) [Seibert et al., 2009]:

display math(2)

[20] With L being the load of the solute and crunoff being the concentration of the solute in stream water. The analytical solution of the integral is then given as (equation ((3))) [Seibert et al., 2009; Winterdahl et al., 2011a]:

display math(3)

[21] Introducing a lower limit to the active soil profile, further referred to as the base depth, d, and a base concentration, cd, for the solute at d, c0 can be calculated as c0 = cde− f(d).

[22] Model calibration of RIM was performed by manual parameter choice from measured soil properties (Table 1) except for the shape parameter f. Transmissivity profiles (parameters a and b) were estimated from a one year record (2009) of stream discharge to groundwater level relationships measured at 11, 6, and 4 locations within the CC-4, NO-5, and NR-7 catchments, respectively. These sites have a distance from the stream of 0.2 m to 60 m (see Figure 1). Base concentrations of the simulated soil DOC concentration profiles were set to 5 mg L−1, which was nearly equal to the lowest measured DOC concentration in stream water of 4.0, 4.7, and 4.2 mg L−1, for CC-4, NO-5, and NR-7 during 2004–2011, respectively. Base depth was set to −1 m for all three streams. Values for d commonly used in the region are −1 m to −2 m; common values used for cd are 5 mg L−1 [Seibert et al., 2009; Winterdahl et al., 2011a]. The shape parameter f was fitted individually for each catchment (CC-4, NO-5, and NR7, respectively) using an unconstrained nonlinear optimization algorithm (“fminsearch” function; MATLAB 7.8.) for minimizing the sum of squared errors (SSE) of modeled versus measured pretreatment stream DOC concentrations (years 2004–2005).

3 Results

3.1 Changes in Soil Temperature and Soil Moisture

[23] Soil temperatures at all soil depths were generally more variable in the clear-cut than in the reference catchment. At 50 cm soil depth Ts varied between 0.6 and 15.5°C (median = 3.4°C) in the clear-cut catchment, whereas values of 0.9 to 12.5°C (median = 3.9°C) were measured in the reference catchment (Figure 2). Highest soil temperatures were commonly observed during late July to early August in both the clear-cut and the control. Further, differences in soil temperatures (∆Ts) at 50 cm soil depth showed a large variation over the year (Figure 2). ∆Ts was small during winter with a stable snowpack, but turned negative (typical value = −0.5 to −1.5°C, i.e., clear-cut colder than the reference site) at the onset of snowmelt in early spring. During summer, the clear-cut was warmer than the control forest with ∆Ts values often reaching more than 4.0°C at 50 cm soil depth. In fall, when air temperatures declined, ∆Ts turned negative (commonly during mid-September), indicating that the soils in the clear-cut catchment cooled down faster when air temperatures decreased. These negative differences persisted until February (Figure 2). Volumetric soil moisture, θ, at 50 cm soil depth varied between 21.4% and 43.9% in the clear-cut and 24.4% and 34.8% in the reference catchment during the 3 year period (Figure 2). Differences in θ (∆θ) were largest during the spring freshet, often reaching 10–12%, but persisted through the summer and fall, indicating that the clear-cut was continuously wetter in lower soil layers during these seasons. At the soil surface (20 cm depth), however, ∆θ declined after longer warm and dry periods (>14 days without precipitation; data not shown), indicating the drier topsoil in the clear-cut relative to the reference catchment during long, dry summer conditions.

Figure 2.

Forcing of catchment soils as a result of forest harvesting in the Balsjö paired catchment experiment. The upper (first, (a)) panel shows the variation of soil temperatures (Ts) at 50 cm soil depth (dotted = harvest; solid = reference catchment); the second panel (b) shows the difference in soil temperatures (∆Ts) of the two catchments. The third panel (c) presents volumetric soil moisture (θ) at the same soil depth (50 cm; dotted = harvest; solid = reference catchment) followed by the differences in soil moisture (∆θ) at 50 cm depth in the fourth panel (d). Data points represent 60 min averages calculated from 5 min values. Gaps are due to temporary malfunctioning of the measurement system.

3.2 Regression Models for Organic Carbon in Stream Water

[24] Simple linear regression models showed that log-transformed discharge (logQ) as well as GW levels at the close and more distant locations, respectively, were all positively correlated with [DOC] in stream water in both the clear-cut and the forest catchment during winter/spring and the summer/fall season, respectively. In contrast, the correlations found for soil temperature at 15 cm, 30 cm, and 50 cm depth, respectively, were negative during winter/spring (Figure 3), but positive during the summer/fall (see supporting information). This pattern was observed for all three catchments. The closer analysis of the drivers of DOC concentrations in stream water in 2009 using the PCA confirmed that shallow GW-levels as well as logQ were strongly correlated with [DOC] according to both PCA components (F1, explaining between 61 and 69% of the variance; F2, explaining 17–28% of the variance; see Figure 4) in both catchments, the clear-cut and the reference catchment, as well as during both seasons, winter/spring and summer/fall. This indicates the dominant role of lateral water transport from riparian wetlands soils into the stream for determining DOC concentrations in stream water. Further, the PCA results confirmed that the role of soil temperatures is opposing in the winter/spring season as compared to the summer/fall season. During winter/spring, increasing soil temperatures were paralleled with decreasing stream DOC concentrations (component F1, but positive F2), whereas increasing soil temperatures during summer/fall were associated with higher stream DOC levels according to PCA component F1, but negative with respect to the second PCA component F2. This pattern was prevalent in both, the clear-cut (CC-4) and the control catchment (NR-7).

Figure 3.

Linear regression models for the clear-cut catchment (CC-4) during winter/spring 2009. Whereas correlations between stream water DOC concentrations (mg L−1) are positive to log discharge (LogQ, mm) and shallow groundwater (GW, in cm) levels at a close and more distant location, respectively, soil temperatures (Ts, °C) at different depth were found to be negatively correlated to DOC concentrations during winter/spring. Volumetric soil moisture (Soil Moist.) at different depth was positively correlated to DOC during this time of the year.

Figure 4.

Principle Component Analysis (PCA) biplots showing the varying drivers of DOC concentrations in stream runoff during the (upper) winter/spring season and the (lower) summer/fall season in 2009 in the reference catchment (left) and the clear-cut catchment (right), respectively. x axis represents the first PCA component (F1) explaining between 61 and 69% of the variance; the y axis represents the second PCA component (F2) explaining between 17 and 28% of the variance. Squares mark soil temperatures at different depth, triangles volumetric soil moisture at different depth; open, dotted circles shallow GW levels as well as LogQ. DOC concentrations are shown as a filled dot.

3.3 Evaluation of Changing Flow Pathways Using RIM

[25] The static RIM model was able to simulate the first-order control of changing flow pathways on the variation of DOC concentrations in stream water. During the pretreatment period (Figure 5), correlation coefficients between measured and modeled DOC were r2 = 0.52, r2 = 0.73, and r2 = 0.41 in the CC-4, NO-5, and NR-7 catchments, respectively (equations for CC-4 are given in Figure 5). More importantly, correlations between DOC residuals (measured [DOC] minus modeled [DOC]) and Q were no longer significant (see Figure 5, panel F for CC-4, as well as supplementary for NO-5 and NR-7), indicating that the important hydrological effect of DOC mobilization by rising riparian GW levels had been removed by applying RIM. However, comparing the measured and modeled DOC concentrations during the entire study period, a strong seasonal signal of deviating concentrations became evident (Figure 5c). This signal was largest during warm summer conditions and resulted in less well-defined relationships between modeled and measured DOC concentrations for the entire 2004–2011 period (r2 = 0.39, r2 = 0.31, and r2 = 0.23 in CC-4, NO-5, and NR-7, respectively).

Figure 5.

First Panel (a): measured (black squares) and modeled (open circles) DOC concentrations of the CC-4 clear-cut catchment using the static RIM model. Second Panel (b): Stream Discharge of CC-4. Third Panel (c): residual (modeled minus measured) DOC concentrations of the RIM model. Fourth Panel: Correlation of measured vs. modeled DOC in the pretreatment period used for model calibration (Panel d) as well as for the entire data set (Panel e). Last Panel (f): relationship between measured discharge (Q) and DOC residuals. Straight lines indicate significant linear regressions. Equations are given in the respective graphs.

[26] When comparing the sensitivity of the DOC residuals to soil temperatures at different depths, relationships were observed (Figure 6). Linear regressions ranged from values around r2 = 0.17–0.20 (all p < 0.001) at a shallow soil depth of 15 cm, to r2 = 0.20–0.23 (all p < 0.001) at 30 cm soil depth to r2 = 0.21–0.31 (all p < 0.001) at 50 cm depth in the clear-cut and the reference catchment, respectively (Figures 6a, 6b, and 6c). Further, the linear models of residuals versus Ts were not found to be significantly different between clear-cut and reference catchment at any given depth (p < 0.05). The linear relationships between residuals and Ts were further supported by positive Pearson's correlation coefficients ranging from 0.42 to 0.55 (all p < 0.001) as well as positive Spearman rank correlation coefficients between 0.40 and 0.55 (all p < 0.001) for the clear-cut and the reference catchment and all three different soil depths, respectively. In contrast, relationships of DOC residuals to soil moisture content indicated a nonlinear behavior (Figures 6d and 6e). This observation was further supported by the fact that no significant correlations could be established using both the Pearson correlation and the Spearman rank test.

Figure 6.

Residuals of DOC (measured minus modeled) of the static RIM model versus soil temperature (Ts) and volumetric soil moisture (θ) at different soil depths: (a) residuals versus Ts at 15 cm depth; (b) residuals versus Ts at 30 cm depth; (c) residuals versus Ts at 50 cm depth; (d) residuals versus θ at 20 cm depth; (e) residuals versus θ at 50 cm depth. Open circles mark data from the reference catchment (NR-7), whereas filled circles represent data from the CC-4 clear-cut catchment. Lines represent significant (p < 0.05) linear regression models (solid line = CC-4; dashed line = NR-7). Equations are given within the graphs (upper equation = CC-4; lower NR-7). Linear regression models for CC-4 and NR-7 were not significantly different in Figures 6a, 6b, and 6c.

4 Discussion

4.1 Hydrological Controls of DOC Mobilization

[27] Enhanced stream discharge during summer following forest harvesting is commonly observed in paired catchment experiments [Andréassian, 2004; Bosch and Hewlett, 1982; Hornbeck et al., 1997]. Earlier work in Balsjö has shown that specific discharge (water volume per unit time and unit catchment area) was increased from 289 mm a−1 in the NR-7 catchment to 487 mm a−1 in the CC-4 catchment during the first 5 years (2007–2011) after clear-cutting [Schelker et al., 2013]. These differences in discharge were largest during summer low flow conditions, when transpiration in the forest was highest [Sørensen et al., 2009].

[28] When considering the strong relationships of stream flow and shallow GW levels that are observed in the region [Bishop et al., 2004] and were also found in Balsjö (see Table 1, supporting information), the differences in discharge can be directly transferred into increased shallow GW levels in clear-cuts compared to forests (see Figure 7 for a conceptual model). Furthermore, there was no indication that this relationship was changed by the performed harvests. This is evident from the fact that the two parameters describing the transmissivity profile of the riparian soils, a and b, were not significantly different among the sites following the harvesting gradient (Table 1).

Figure 7.

Conceptual model of the controls of DOC mobilization in a (left) clear-cut and an (right) undisturbed boreal first-order catchment. Figure 7a1 shows the increased hydro-climatic forcing of DOC source areas (light grey) as compared to the control forest shown in Figure 7a2. The first-order hydrological control of stream DOC variations as modeled with the RIM approach is shown in Figure 7b1 for the clear-cut and in Figure 7b2 for the forest catchment. This control is based on increased shallow GW levels in the clear-cut as compared to the control forest that in turn flush more surficial soil layers that have higher DOC concentrations. Water flows are shown as black arrows; data points of total organic carbon (TOC) in shallow GW represent measured concentrations in Balsjö [modified from Schelker et al., 2012]. The second-order soil temperature control of DOC in stream water is represented in Figures 7c1 and 7c2 for the clear-cut and the forest catchment, respectively. The soil temperature range marked as (I) shows the typical variation during the spring freshet where stream DOC is controlled by various winter and spring processes. The range marked as (II) indicates summer conditions where the clear-cut reaches significantly higher soil temperatures than the control forest.

[29] Increased GW levels were further found to enhance the transport of DOC into streams (Figures 7b1 and 7b2). Support for this finding can be derived from multiple lines of evidence. First, the strong correlations between shallow GW levels and stream DOC concentrations (Figure 3; Figures A1, A2, and A3, supporting information1) clearly indicate that an activation of shallow soil layers increased DOC concentrations in the streams. Interestingly, this correlation was better (r2 = 0.76 versus r2 = 0.60) for the more distant riparian GW tube (distance = 25–35 m from stream; see also section 2.2.) in CC-4 than for the near-stream (<0.5 m from the stream) one, which may simply be a result of the fact that the near-stream GW levels may not only be affected by lateral water flows from uplands to streams, but also by hyporheic exchange. Second, both the strong correlation of shallow GW levels and DOC in streams, as well as the slightly better correlation of GW level versus DOC in the more distant location as compared to the near-stream one were also confirmed by the results of the PCA. More specifically, the results of the PCA show the dominant role of shallow GW levels in defining stream DOC concentrations. DOC was closely associated with shallow GW levels during different seasons (winter/spring vs. summer/fall), with and without harvesting effect, with the exception of the winter/spring season in the control forest where soil moisture was a better predictor than shallow GW level (Figure 4).

[30] Furthermore, an exponential decrease of DOC concentrations with soil depth in riparian soils is commonly found in the region [Bishop et al., 2004; Grabs et al., 2012; Laudon et al., 2011] and has also been observed during one sampling campaign in Balsjö in August 2009 [Schelker et al., 2012]. Near the soil surface, these concentrations were nearly five times higher than the ones observed in the stream (134.7 mg L−1 at 20 cm soil depth versus 28.1 mg L−1 in the stream) and the concentrations at 90 cm soil depth (24.7 mg L−1; Figure 7) were almost equal to the ones in the stream. These results reaffirm the conceptual model that organic carbon is mobilized from peat-rich riparian soils when shallow GW levels rise (Figures 7b1 and 7b2). Finally, additional support for this control can be found in the fact that the static-RIM model was able to explain between 41% and 73% of the total variation in DOC concentrations during the pretreatment period by mimicking the variations in shallow GW levels. The results of this study therefore suggest that shallow flow pathways governed by increasing water flows through catchment soils represent a first-order control on regulating DOC concentrations in boreal first-order streams that are subject to forestry operations.

4.2 Soil Temperature Effects on DOC Availability in Riparian Soils

[31] A soil temperature control on stream DOC concentrations has been suggested earlier [Freeman et al., 2001; Worrall et al., 2004]. In addition, many models for terrestrial and aquatic C-cycling assume a temperature-dependent production of DOC in catchment soils, often conceptualized as a constant fraction of solid OM decomposition or as a direct function of temperature using a rate constant [Futter et al., 2007; Svensson et al., 2008] which is similar to the Q10 rate constant for simulating OM decomposition in many terrestrial C-models [Ågren and Bosatta, 1996; Conant et al., 2011; Kirschbaum, 1995]. However, this hypothesis has been discussed, and a laboratory study has shown that increases in soil water DOC concentrations were not a direct result of soil warming, but rather the season [Jones and Willett, 2006]. Furthermore, several studies have suggested a temperature-dependent accumulation of DOC in catchment soils that is then flushed into surface waters during events [Boyer et al., 1997; Hornberger et al., 1994; Roulet and Moore, 2006], which would lead to a nonlinear effect of soil temperature on stream DOC concentrations. More specifically, the varying DOC availability in soils could cause a hysteresis of DOC concentrations when DOC pools are diminishing during larger runoff events.

[32] In Balsjö, soil temperatures were observed to increase up to +5°C at 50 cm soil depth in the clear-cut compared to the reference catchment during summer (Figure 2). These increases were correlated with DOC residuals (Figures 6a, 6b, and 6c) indicating that higher soil temperatures increased the availability of DOC in riparian soils (Figures 7c1 and 7c2). These findings agree with recent studies highlighting the dependence of stream water DOC concentrations on soil temperature during summer as a second-order control [Köhler et al., 2009; Winterdahl et al., 2011a; Winterdahl et al., 2011b]. Considering the fact that linear regression models for Ts versus DOC residuals of this study were not significantly different between the clear-cut and the reference catchment at any given soil depth (Figures 6a, 6b, and 6c), this study suggests that the warming induced increase in DOC availability in riparian soils is not different in the clear-cut compared to the control forest. In addition, this finding suggests that even if water flows through catchment soils are enhanced by 198 mm a−1 after clear-cutting (see section 4.1.), the DOC export from clear-cut soils does not become supply limited and does therefore not show particular signs of hysteresis during summer.

[33] Overall this study suggests that the regulation of stream water DOC during summer is soil temperature dependent. To our knowledge this study is the first field-scale experiment that has been able to show this link between soil warming as a result of the performed experimental harvests and increased DOC mobilization into streams.

[34] Understanding the mechanisms that cause this temperature dependence of DOC availability in the catchment soils is challenging. First, it should be noted that a major part of the production of DOC in catchment soils may be lost by C mineralization before it can be transported to the stream channel, so that the “net-production” of DOC for the transport into surface waters is the difference between DOC production and C mineralization. A question arising from the dependence of the net-production of DOC on soil temperature is why DOC concentrations increase in the first place and are not compensated by increased C-mineralization in catchment soils, if both these processes are temperature dependent. Fast turnover of DOC to CO2 has been hypothesized as a mechanism compensating for the increased leaching of DOC in soil water of temperate forest ecosystems that were subject to forest harvesting [Bengtson and Bengtsson, 2007]. Further, this hypothesis suggests that the turnover rate of organic carbon would be equal or higher to that of DOC production so that the change in net-production would be negative or zero. In contrast, our results suggest that the temperature dependence of C-mineralization would be lower than the one for DOC production, so that the change of the net-production of DOC is positive and creates a surplus that is then available for hydrological mobilization into surface waters.

[35] During the spring (marked as I in Figures 7c1 and 7c2), the factors controlling DOC besides the hydrological control were less clearly explained. This additional variation may be related to the dynamic nature of the spring freshet in the region [Laudon et al., 2002]. This dynamic may be even more pronounced after forest harvesting with higher runoff and different timings of peak flows in clear-cuts compared to the control forest [Schelker et al., 2013; Sørensen et al., 2009]. Furthermore, recent studies have highlighted the importance of winter conditions and specifically soil frost for stream DOC concentrations during the spring [Ågren et al., 2010; Haei et al., 2010]. We propose therefore that spring DOC levels after clear-cutting are not dependent on soil temperature, but instead on preceding winter conditions. One indication for this proposed seasonal change in controlling factors may be found in the fact that DOC concentrations during spring were negatively correlated to increasing soil temperatures (Figures 3d, 3e, and 3f). This indicates that higher soil temperatures during the spring cause an increased melt rate. The increased melt water contributions to the stream may then dilute the DOC concentration and therefore partly offset the increasing DOC concentrations caused by increased GW levels during the spring freshet.

4.3 Soil Moisture Effects on DOC Availability

[36] Model conceptualizations for OM decomposition (and DOC production) commonly assume a negative parabolic curve for the dependence of microbial decomposer activity on soil moisture. The function defines optimum soil moisture conditions at which the microbial activity is highest resulting in lower activity in both dryer and wetter soils [Ågren and Bosatta, 1996; Futter et al., 2007; Svensson et al., 2008].

[37] The nonlinear behavior between model residuals and θ found in this study (Figures 6d and 6e) do in principle support such conceptualizations of a soil moisture optimum. Assuming that the model residuals represent the rate of net-production of DOC in soils (i.e., no additional processes such as for example sorption of DOC on mineral surfaces are considered), an optimum net-production of DOC at an intermediate soil moisture state could be indicated by the data. However, the strong scatter in both soil moisture plots (Figures 6d and 6e) makes it difficult to draw any direct conclusion from this nonlinear pattern in the data.

4.4 Study Limitations

[38] One aspect that may limit the results of this study is the question whether additional C-inputs from logging residues to the soil surface have an effect on stream DOC levels after clear-cutting. In the Balsjö catchment, logging residues (tree tops, branches with needles and stumps) were left on site [Löfgren et al., 2009] and could therefore have contributed to the mobile OM pool. However, given that the C pool size of logging residues is small compared to the soil C pools of riparian areas that can have peat layers that are more than 2 m thick (see section 2.1.), we considered their contribution to be minor. This assumption is also in accordance to earlier work on DOC in the Balsjö catchment that has assumed a negligible role of logging residues on the dynamics of stream DOC [Laudon et al., 2009; Schelker et al., 2012].

[39] A second limitation of the results of this study is given by the structure of the modeling approach. In principle, the model residuals that are used to identify higher-order controls are not only a summary of all the remaining variance (for all higher-order controls) in the data set that are not yet explained by the model, but also the uncertainty of the data itself (measurement uncertainty), as well as the uncertainty of the model structure and parameterization used to simulate the lower order controls. Given that we used the RIM approach to model the first-order hydrological control, the fundamental RIM assumptions become critical. These assumptions are [Grabs et al., 2012] (i) an exponential decrease of transmissivity with soil depth, (ii) exponentially shaped soil solution profiles, (iii) negligible lateral flows in the unsaturated zone, as well as below d, (iv) 100% soil matrix flow (no overland or macropore flow) and time invariant concentrations of cd. Even if we did not find any evidence for a direct violation of any of these assumptions, we note that a violation would become critical for the further interpretation of the model residuals and for evaluating the role of higher-order controls.

5 Conclusions

[40] In summary, the results of this study indicate that increased concentrations of DOC in boreal first-order streams following forestry operations are primarily controlled by the increased flux of water through shallow flow paths following forest harvesting. Increases in soil temperatures as a result of clear-cutting appear to be a second-order control on [DOC] during summer because it increased the availability of DOC in riparian soils. The mechanisms of DOC mobilization remain therefore the same as in undisturbed forest ecosystems, but are amplified in the clear-cut areas. According to our observations, the hypothesis that these drivers of DOC mobilization in clear-cuts may be the same as in undisturbed forest catchments was not falsified. Further, our study provided an explicit test of the “soil temperature DOC-feedback hypothesis.” The evidence we found from the altered hydro-meteorological forcing of catchment soils after forest disturbance strongly support this hypothesis, at least during summer. However, the large control of flow pathways on DOC concentrations suggests a higher sensitivity of boreal streams to changes in precipitation and transpiration controls on shallow groundwater flux, than increased soil warming during summer. These observations further indicate that boreal streams are vulnerable to both land use and climate change.

Acknowledgments

[41] Funding for this work was provided by CMF, Future Forests, and the Formas (ForWater). We thank Ida Taberman, Viktor Sjöblom, and Peder Blomkvist for help in the field and the laboratory. Further, we would like to thank the two anonymous reviewers for their constructive comments that have largely increased the quality of this paper.

Ancillary