Climate change projections show that temperature and precipitation increases can alter the exchange of greenhouse gases between the atmosphere and high latitude landscapes, including their freshwaters. Dissolved organic carbon (DOC) plays an important role in greenhouse gas emissions, but the impact of catchment productivity on DOC release to subarctic waters remains poorly known, especially at regional scales. We test the hypothesis that increased terrestrial productivity, as indicated by the normalized difference vegetation index (NDVI), generates higher stream DOC concentrations in the Stordalen catchment in subarctic Sweden. Furthermore, we aimed to determine the degree to which other generic catchment properties (elevation, slope) explain DOC concentration, and whether or not land cover variables representing the local vegetation type (e.g., mire, forest) need to be included to obtain adequate predictive models for DOC delivered into rivers. We show that the land cover type, especially the proportion of mire, played a dominant role in the catchment’s release of DOC, while NDVI, slope, and elevation were supporting predictor variables. The NDVI as a single predictor showed weak and inconsistent relationships to DOC concentrations in recipient waters, yet NDVI was a significant positive regulator of DOC in multiple regression models that included land cover variables. Our study illustrates that vegetation type exerts primary control in DOC regulation in Stordalen, while productivity (NDVI) is of secondary importance. Thus, predictive multiple linear regression models for DOC can be utilized combining these different types of explanatory variables.
Citation: Mzobe P, Berggren M, Pilesjö P, Lundin E, Olefeldt D, Roulet NT, et al. (2018) Dissolved organic carbon in streams within a subarctic catchment analysed using a GIS/remote sensing approach. PLoS ONE 13(7): e0199608. https://doi.org/10.1371/journal.pone.0199608
Editor: Andrés Viña, Michigan State University, UNITED STATES
Received: September 8, 2017; Accepted: June 11, 2018; Published: July 6, 2018
Copyright: © 2018 Mzobe et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and Supporting Information files. The data is also available from Dryad at the following link: https://doi.org/10.5061/dryad.pq788v9.
Funding: The EUROSA Project (Erasmus Mundus Action 2 Partnerships scholarship programme) provided PM with an exchange scholarship at Lund University. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
In subarctic and arctic regions, much of the carbon is in the form of soil organic carbon (SOC), in soil that is partly frozen (permafrost) . Permafrost accounts for 24% of the land area at northern latitudes . The global reserve of permanently frozen SOC is estimated at 1700 Pg C, twice that of the atmosphere . A number of climate models predict that northern regions (above 60°N) will experience climatic changes, in particular warmer temperatures and higher precipitation [4, 5]. These changes have environmental effects such as permafrost thaw , altered hydrology , erosion of palsa mires , vegetation change such as tree line and shrub expansion , and increasing organic matter loading, especially as dissolved organic carbon (DOC) in rivers .
An implication of permafrost thaw is that the active layer in soil deepens, such that previously frozen SOC becomes susceptible to microbial degradation, which releases CO2 and CH4 [11, 12]. Moreover, thawing permafrost promotes the release of old carbon into rivers as DOC , where it can be further degraded into CO2. This lateral transport of carbon via DOC could be expected to further increase in future climate change scenarios of increasing precipitation and increased hydrological flows [14, 15].
Climate warming causes phenological changes such as a shift in the growing season start and end (extended span), an increase in net primary productivity (NPP), and increased forest cover [16, 17]. This increase in biomass allows terrestrially derived organic matter to become an increasingly important lateral source of carbon at the catchment scale. Therefore, understanding the release of DOC from northern catchments, which have a store of permafrost, together with increased terrestrial productivity is a major concern in biogeochemical research.
Strong evidence for altered land-water coupling in the carbon cycle at northern latitudes e.g. Scandinavia, Canada, and Germany is the phenomenon of brownification of inland waters [18–20]. The change in water colour to brown is mainly due to an increase in terrestrially-derived dissolved humic substances , typically being reflected in increased DOC concentrations [22–24]. Other causes of brownification that have been suggested in the literature include changes in vegetation, declining sulphate deposition, an increase in precipitation, and land use change [25–28]. Browner waters limit light penetration in water bodies, thus decreasing the primary production and affecting secondary production of aquatic fauna [29, 30]. Brownification also leads to increased remineralization of DOC in inland waters, and thus increased fluxes of CO2 to the atmosphere [31, 32].
Although the contributions from inland waters to gaseous carbon in the atmosphere are increasingly acknowledged in global carbon cycle studies, the lateral export of dissolved organic carbon (DOC) at high latitude regions remains an emerging area of research . Lateral interactions, e.g. permafrost thaw, have the potential to increase the carbon release to the atmosphere from high latitude regions, possibly turning landscapes from carbon sinks into sources [34, 35]. Further research is thus needed to reach a conceptual understanding of the controls on DOC release from catchments, particularly in subarctic regions which are strongly affected by climate change. One way to do this is to investigate the coupling of landscape components in biogeochemical cycles such as carbon at varying spatio-temporal scales [36, 37]. Another step is to refine regional estimates using direct measurements and topographical data. Research in northern Sweden has begun to do this by incorporating lateral carbon fluxes in models through the use of distributed hydrological modeling, and mixing-models using geological landscape elements [38, 39].
Multiple studies in Fennoscandia have used the Normalized Difference Vegetation Index (NDVI) to quantify changes in vegetation productivity and linked them to DOC concentration in recipient waters [27, 39, 40]. Larsen et al.  used the NDVI as a proxy for vegetation density in their study of lake organic carbon concentrations in Norway. They found that vegetation density was a strong positive predictor of lake organic carbon concentrations. Moreover, in a continental (Europe) scale study of soil pore waters, a positive correlation between the NDVI and DOC concentration in broadleaf forests was found . However, there is need for further studies to understand how NDVI and other catchment properties regulate DOC in the subarctic, and whether or not a priori knowledge about land cover type (mire, forest etc.) needs to be included in order to build a predictive model for sub-catchment DOC variations.
In this study we aim to determine (a) the relationship between the NDVI and DOC at subcatchment level, (b) the degree to which other generic catchment properties (elevation, slope) explain DOC, and; (c) whether or not additional land cover variables representing the local vegetation (mire, forest, lake, tundra) need to be included in predictive models for DOC delivered into rivers. We hypothesize that subcatchments with higher terrestrial productivity generate higher DOC concentrations in the subarctic catchment Stordalen, Sweden. Temporally, we anticipate a time lag in the DOC response to changing productivity. This arises as we assume that the effect of productivity during a given year will be captured by DOC measurements performed throughout the growing season and subsequent period, ending at early snowmelt of the following year.
2. Materials and methods
2.1. Ethics statement
Data used in this study was collected with permission from the Abisko Scientific Research Station (ANS) by Olefeldt et al.  and Lundin et al.  who supplied us with the data for Stordalen. Endangered or protected species were not used in this research.
2.2. Site description and experimental design
Stordalen catchment (15 km2; 68°21’N, 19°03’E) is located in northern Sweden, 10 km east of the village of Abisko (see Fig 1). It lies in the subarctic climatic zone with the coldest temperatures in February and the warmest in July . Mean annual temperature (MAT) for the period 2000 to 2009 was between 0.6° and 0.7°C [46, 44]. This is higher than at the start of the 20th to early 21st century, where MAT temperatures were between -0.7°C and -0.6°C [47, 48]. Mean annual precipitation lies between 304 mm (in 2003) to 308 mm (in 2006) [47, 34].
The topography in Stordalen is characterized by a gradient from high elevation (≈800 m.a.s.l.) in the southern part to low lying topography (≈360 m.a.s.l.) in the north. The upper (southern) catchment has steep slopes consisting of igneous and metamorphic rock outcrops and tundra, with patchy lakes and mires. The streams in this part of the catchment are single channels which are fed by precipitation and snowmelt. Lower catchment streams are dispersed and drain into larger lower lying lakes. Abundance of lakes and peatland mires in the lower part of the catchment is facilitated by gentle slopes, moraine and peat soils. The inflection point of the catchment is marked by birch forest (Betula pubescens spp) and further towards the north, by the emergence of mires and lakes .
Stordalen mire is characterized by three land cover types: palsa, sphagnum, and eriophorum, respectively . These land cover types are found where permafrost degradation has taken place in the catchment.
We selected ten nested subcatchments, defined by 10 monitoring points in Stordalen catchment, to achieve the aims of this study. Each subcatchment has a mixture of land cover types i.e. mire, birch forest, lakes etc. shown in Fig 1. First, single relationships between monthly averaged DOC concentration and variables that reflect catchment productivity and main geographical features (slope, elevation) were tested. We then built multiple linear mixed effects regression (LMER) models to explore how DOC predictability might be improved when land cover variables are included. Finally, results from the LMER models were extrapolated and applied to the whole Stordalen catchment, in order to visualize the contribution to surface water DOC from each pixel in the catchment.
To quantify the vegetation productivity we used the NDVI as a proxy for net primary production (NPP) i.e. vegetation productivity. The NDVI has been used to represent NPP in the field [51–54], particularly in Light/Rain Use Efficiency (L/RUE) models based on the Eq 1 below from Yengoh et al. : (1)
We used the NDVI from the easily accessible Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI data as an indicator of NPP in Stordalen.The NDVI is a widely used and relevant indicator used in a plethora of land use change studies and finds use in a number of northern latitude studies e.g. . It was selected as a quick broad scale assessment of terrestrial productivity.
2.3. Field sampling and laboratory analysis
Water monitoring data was obtained for 2007 and 2008 from Olefeldt et al. [43,56]. For the period 2009 to 2011 data was provided by Lundin et al. . Stream grab samples together with samples from in situ autosamplers were taken and analyzed for DOC (in mg L-1), water temperature (°C), pH, total nitrogen, and conductivity. Water samples were taken at a monthly interval using an acid washed polythene bottle which was refrigerated prior to analysis . The autosamplers (Hach Sigma 900) collected water on a 3 to 6 days cycle . The samples underwent filtration using a 0.45 μm filter followed by acidification with HCl. The samples were then inserted in the Shimadzu TOC-V CPH analyser to determine the DOC content . We utilized the 10 monitoring points by Olefeldt et al.  as a basis for sub-catchment delineation, as shown in Fig 2.
Stream stage was measured 2007–2008 at hourly intervals using Odyssey capacitance water level probes . Hourly direct discharge (m3/s) measurements were approximated using the velocity-area method, using an Aqua Data Sensa Electromagnetic Current Meter to measure velocity . Stream stage and discharge data were used to generate rating curves which were then used to derive hourly discharge for the snow free period . Since discharge data was available for two years only (2007–2008) and since the number of DOC measurements varied between sites and years, complete DOC exports were not possible to calculate in this study. Therefore, the DOC concentration was used as the main carbon release indicator in the catchment as there was a longer record of concentration data than of discharge data.
To analyze relationships between DOC and single catchment property variables (aims [a] and [b] in Section 1), the discharge weighted DOC (DOCqwc) was used as a response variable  providing an integrated representation of the overall DOC delivery to recipient downstream waterbodies. The DOCqwc was calculated for each subcatchment based on DOC concentration (C) and discharge (Q) (Eq 2) for the period 2007–2008 when flow data was available.(2)
The DOCqwc provides a representation of the mean DOC concentration in all runoff that is received by downstream water bodies such as lakes . Assuming that areal discharge is similar for the different sub-catchments, DOCqwc also serves as a proxy for spatial differences in total DOC export.
2.4. Data preparation
Elevation data were obtained from the Swedish National Land Survey (Lantmäteriet). The 2 m GSD-Höjddata, grid 2+ is a publicly available, licensed, ASCII format, DEM downloaded in the projection SWEREF99 . Using the ArcMap® Hydrology Toolbox, rivers and subcatchments were assigned using the single flow direction (SFD) D8 algorithm. Artefacts in the landscape such as sinks and anthropogenic structures, e.g. roads and railways, were breached to replicate the natural flow of water within the catchment.
The moderate-resolution imaging spectroradiometer (MODIS) was used as it provided NDVI (MOD13Q1) 16 day composites at 250 m resolution. MODIS data was downloaded for the years 2006 to 2012 from the Oakridge National Laboratory Distributed Archive Center . The data was downloaded in the geographic coordinate system WGS84 and projected to SWEREF99. The NDVI is a unit less vegetation index values ranging between -1.0 and 1.0 (maximum greenness) using the general equation (Eq 3) below.(3)
The annual seasonal mean NDVI of each subcatchment was extracted using Zonal Statistics as Table in Spatial Analyst Tools in ArcMap 10.2.2. This process was repeated for each NDVI TIFF file for the period 2007–2011, for the months of May to September, to yield the mean of the two 16 day composite NDVI per subcatchment and month. The Slope tool (ArcGIS) was used to extract the degree of slope for the catchments from the DEM. Mean slope for each subcatchment was again extracted. Mean elevation (m.a.s.l) and land cover classes (mire, forest, tundra, slope, elevation)  per subcatchment were extracted using the same zonal statistics method.
2.5. Statistical analysis
The correlation between the NDVI, DOC (arithmetic mean per year), and DOCqwc (flow-weighted mean per year) was tested using simple linear regression . Relationships between DOC and additional catchment variables (NDVI, mire, forest, lake, tundra, slope, and elevation) were tested in the same way using SPSS. To eliminate redundancy in input variables, the tundra class was removed since tundra cover is equal to the area not covered by mire, forest or lake.
A weakness that arises with calculating individual annual mean DOC values for the 10 respective subcatchments is that the number of data points becomes low, leading to low statistical power. Hence, in the final analysis part (aim [c]), more robust predictive models for DOC (LMER) was applied by using DOC data for multiple time points per year (monthly averages). LMER models for DOC were built using the statistical package “lm4” in R® with NDVI, slope, elevation and land cover (forest, mire, lake) as predictor variables, and time (months since start of measurements) as random effect variables. The annual mean NDVI was extracted per subcatchment for the period May to September. The percentage of land cover per subcatchment was calculated and used to determine the cumulative contribution into or from a subcatchment, e.g. subcatchment A1 receives the sum of all subcatchments as it is the outlet for Stordalen. This approach ensured that the land cover in the input tables for the LMER was representative of the class for the catchment and not just the subcatchment level.
A LMER model was first established by entering all explanatory variables. Then backwards elimination was carried out by manually removing one explanatory variable at a time (in all possible combinations) based on the p-value, to determine whether or not it could be removed without significantly changing the model and losing predictability. An ANOVA test was performed at each exclusion step for cross model comparison. To obtain conditional (complete model) and marginal R2 estimates (R2c and R2m, respectively), the package MuMIn was used, while significance of coefficients (fixed effects) were tested with the package lmerTest. The results of the LMER models were classified into six classes using the geometrical interval in ArcGIS.
Ten subcatchments were delineated. The subcatchments had good representation of the main flow paths and associated lakes in Stordalen (Fig 2). A distinct pattern of lakes at lower elevations, i.e. below the catchment inflection point, was observed with rivers at higher elevations typified by straight and narrow channels.
3.1. NDVI-DOC relationship
Peak NDVI values were consistently observed in July as shown in Fig 3. NDVI changes from 2006 to 2012 showed that the eastern periphery and southern subcatchments e.g. A2, B2, A7, and C1 experienced the most change in NDVI. Interior subcatchments showed little change in annual comparisons.
Annual mean NDVI for the catchment is shown with subcatchment boundaries in black overlain to annual variation across subcatchments.
To determine whether the NDVI had an effect on the mean annual DOC concentration (n = 30) and mean annual DOCqwc (n = 20) in the catchment, a simple linear regression for the years 2007–2009 was performed with the significance threshold set at 0.05. The DOC yielded an r-squared of 0.137 (p = .044) for DOC and the DOCqwc an r-squared value of 0.114 (p = .146).
The DOC vs NDVI (Fig 4A) had a statistically significant relationship, whilst the DOCqwc vs NDVI (Fig 4B) did not. However, the trendline for DOCqwc was positive which could have supported our hypothesis had the p-value been lower. Subcatchments with high NDVI yielded low DOC concentration values.
Scatterplot of (a) mean annual dissolved organic carbon (DOC) concentrations and (b) flow-weighted mean annual DOC against NDVI per subcatchment. The DOC regression line shows Y = -41.765x + 41.971, R2 = 0.137 and for DOCqwc Y = 0.0042x + 0.7208, R2 = 0.114.
Simple linear regression was again performed to assess the extent to which slope can explain the observed DOC fluctuations. Results (Fig 5A) yielded a negative significant relationship with an r-squared value of 0.461 (p < .001) for DOC.
Scatterplot of (a) mean annual dissolved organic carbon (DOC) concentrations and (b) flow-weighted mean annual DOC against Slope. The DOC regression line shows Y = -0.5047x + 14.816, R2 = 0.461 and for DOCqwc Y = 0.2789x + 6.9655, R2 = 0.082.
The DOCqwc had a non-significant relationship with slope (Fig 5B), producing an r-squared of 0.082 (p = .220).
The process was repeated to assess influence of elevation on the DOC. The linear regression results for DOC (Fig 6A) yielded a significant negative relationship with an r-squared value of 0.374 (p < .001) for DOC concentration.
Scatterplot of (a) mean annual dissolved organic carbon (DOC) concentrations and (b) flow-weighted mean annual DOC against Elevation per subcatchment. The DOC regression line shows Y = -13.209x + 628.36, R2 = 0.374 and for DOCqwc Y = 8.853x + 408.67, R2 = 0.099.
The DOCqwc had a non-significant relationship with the NDVI (Fig 6B), producing an r-squared of 0.099 (p = .178).
In our second objective we sought to determine the extent to which catchment properties (i.e. land cover) influences DOC in Stordalen. Earlier in Section 3.1 the NDVI, slope, and elevation were compared against DOC and DOCqwc. However, the NDVI, slope and elevation are not representative of all generic (sub) catchment variables e.g. vegetation type. A Pearson’s Correlation was thus performed, testing for significance at the 0.05 and 0.01 levels between land cover classes (mire, forest, lake, tundra, slope and elevation) and DOC for all available years. The results of the correlation (Table 1) show that DOC has positive significant correlations with the mire, NDVI, and forest. The DOC had significant negative correlations with slope and tundra.
Slope (n = 50), Mire (n = 50), Forest (n = 50), Lake (n = 9), NDVI (n = 50), Elevation (n = 50), and DOC (n = 218).
The NDVI had a positive and significant correlation with forest, DOC and mire. Conversely, the NDVI has a negative and significant correlation with slope, elevation and tundra. NDVI values for the years 2007 to 2011 were consistently low for subcatchment B2 which has steep slopes, tundra, rock outcrops, and high elevation values. Subcatchment A2 also had low NDVI values, second to B2, during this period. High NDVI values were found in subcatchments A1 and A5 in the five year period. Subcatchments A6 (2008–2009), A7 (2010) B1 (2007), and C1 (2011) had the third highest NDVI values in their respective years.
Slope had a positive significant correlation with tundra (.907) only. In contrast, tundra showed significant negative correlation with mire, lakes and DOC. Notably; slope had negative and significant relationship with all land cover classes, with the exception of tundra.
Mires had the strongest, positive and significant correlation with lakes, DOC and NDVI. However, they (mires) showed a strong negative correlation with tundra (-.934) and slope (-.923). Lakes had a positive and significant relationship with mires.
3.2. Linear mixed effects models
The backwards elimination procedure resulted in three explanatory models for Stordalen catchment (summarized in Table 2). In all three cases, all of the coefficients were significant (p < 0.05). In the exclusion process, removal of specific land cover variables (i.e. mire, forest, lake), yielded a poorer fit. Hence, these variables were kept in all models. The tundra class was not included in this step as it constitutes what is not covered by mires, forest and lake land cover classes.
Models are ranked by marginal and conditional R squared values. Slope (n = 50), Mire (n = 50), Forest (n = 50), Lake (n = 9), NDVI (n = 50), Elevation (n = 50), and DOC (n = 218).
In all models, the mire class stands out as a positive DOC explanatory variable in Stordalen. In Model 2, the NDVI serves as a secondary significant positive predictor of catchment DOC. With the exception of the full model, Model 2 is the only model that includes the NDVI. Similarly, elevation and slope appear only in Model 3 and Model 1, respectively. The full model, which provides a comparison with the three models, had the lowest marginal and conditional r-squared values and the highest AIC and BIC values.
The significance and correlation tables of the full model were exported to provide the overall understanding of the LMER results (Table 3 and Table 4). Results of the fixed effects (Table 3) showed that significant predictors were slope, mire, lake and forest classes.
Slope (n = 50), Mire (n = 50), Forest (n = 50), Lake (n = 9), NDVI (n = 50), and DOC (n = 218).
Slope (n = 50), Mire (n = 50), Forest (n = 50), Lake (n = 9), NDVI (n = 50), Elevation (n = 50), and DOC (n = 218).
The NDVI and elevation were not significant predictors of DOC in the full model as shown in Table 3. They do emerge in Models 2 and 3 which have higher R2m values when compared to the full model. Lakes had the highest negative trend whilst mires showed the opposite. In the full model, forest and slope terms were negative, whilst the NDVI term was positive. The negative pattern by forest and slope was mantained in Models 1 to 3 with elevation being negative where it appeared as a fixed effect in models.
3.3. Spatial representation
We extrapolated the equatons in Table 2 to each pixel of the study area to visualise what the models suggest about how each part of the Stordalen catchment contributes to surface water DOC.
Model 1 (Fig 7) shows that lakes, found largely in low lying subcatchments A3 to A1, contribute the least to DOC export in Stordalen. Another region were DOC contribution was minimal was where there were steep slopes and a mixture of omitted classes such as tundra and outcrop (e.g. upper B2, A7, A4 and A6).
Coefficients and the intercept (52.9) were extrapolated from LMER Model 1 to each pixel using the fixed terms of the model equation. DOC loss (denoted in blue) occurs in lakes and where steep slopes occur with forest land cover.
Interestingly; locations surrounding mires also had high contributions towards DOC. These areas were characterized by low slope angle and forest cover (e.g. A6 and A5).
In Model 2 (Fig 8) we found that forest cover lowered the effect of high NDVI values with its negative sign in the LMER equation, thus accounting for the large spread of the green (-1 to 12.1) grouping in Fig 8.
Coefficients and the intercept (1.4) were extrapolated from LMER Model 2 to each pixel using the fixed terms of the model equation. DOC loss (denoted in blue) occurs where lakes and forest land cover dominate.
The tundra class (upper regions of subcatchments A6, A7, and B2) was included by omission, i.e. the area not covered by any other land cover classes. It shows up as moderate to low values.
Mire is the only class in Model 3 that does not have a negative coefficient and shows up as the strongest contributor of DOC in the catchment (Fig 9). Model 3 lacks intermediate classes due to the fixed effects having negative values associated with them (mires excluded). This is already evident in Table 2 where the mire class is the only variable with a positive coefficient.
Coefficients and the intercept (17.3) were extrapolated from LMER Model 3 to each pixel using the fixed terms of the model equation. DOC loss (denoted in blue) occurs where lakes and high elevation combine with forest land cover.
Steep to moderately steep areas that have forest cover, shown as the dark blue band in subcatchments A7, A6, A4 and B2, stand out in this model as they together with lakes, have the lowest contribution to DOC.
Our hypothesis was to show that increased terrestrial productivity, represented as the NDVI, generates higher DOC concentrations in the Stordalen catchment. Using simple linear regression in our first objective, the NDVI showed a significant negative relationship with DOC concentration. DOCqwc was marginally significant with the NDVI. This was interpreted as a difference in scale that the two DOC indicators capture. More specifically, the flow-weighted mean DOC better represents DOC concentration in the entire export. Thus, the NDVI did not appear to show total export. At the local level, high DOC values can be found where NDVI values are low. Low NDVI values coincided with land cover classes such as mires. The NDVI had its highest significant positive correlations with mire and forest land cover classes. Birch forest, having the largest land cover type in Stordalen, and being detected by the NDVI, contributes organic matter litter in the subcatchments thereby enriching the DOC pool.
LMER tests were further carried out to tease out the NDVI-DOC relationship in Stordalen adding land cover variables into the analysis. The NDVI had a positive but secondary regulatory effect on DOC as it was a predictor variable in Model 2. Furthermore, the NDVI together with elevation were non-significant in the full model when compared with other fixed effect terms. Mires, in all models, were the strongest positive determinant of DOC. These findings of the NDVI-DOC relationship show that the NDVI on its own is not a good indicator of the impact that terrestrial productivity has on DOC in recipient waters, probably because land cover variables (especially mires) exert the primary control on DOC and our window of the NDVI analysis in terms of years was brief.
Our result that mires are key predictors of DOC rich subcatchments in Stordalen supports the findings of other studies that reported them to have an additive effect and constitute one of the key controls of DOC export and concentration in catchments [61–63]. In the subarctic, it has been found that forests contribute to DOC export but spatially their contribution to DOC export is exceeded by mires . Thus, the presence of mires reduces the NDVI-DOC relationship at the landscape level.
Besides the mire class, other catchment properties such as slope, elevation, tundra, and forest (birch) were compared against DOC. A significant negative relationship between DOC concentration and slope was found. This adheres to catchment land cover distribution in Stordalen as localities where steep slopes dominate, are largely covered by tundra and rock outcrops, while gentle near flat topography is host to mires. This assertion is supported by the significant positive slope and tundra correlation value of 0.907. Slope also had a strong negative, near linear, and significant correlation with the mire class. This reinforced flat areas, which are host to mires, as DOC hotspots in Stordalen. In the LMER models slope was a significant negative and primary predictor of DOC. Low DOC concentration at steeper slopes in the catchment is congruent with hydrological solute transfer by gravity i.e. quick downstream transfer. The positive non-significant relationship DOCqwc had with slope indicates that at the broader scale, catchments that have steep slopes will efficiently transport DOC to a recipient water body at the end of the system. Thus, increasing DOC values with slope can be related to flushing. Slope as a key predictor of DOC is not limited to Stordalen as slope has been observed as a strong predictor at continental level .
Elevation was a tertiary predictor of DOC in the study where it paired best with mire, forest, and lake classes in Model 3. However, as a fixed effect term it was non-significant. In the simple linear regression, elevation had a negative relationship with DOC concentration. This is a function of catchment land cover type, where high elevation zones are characterized by rock outcrops and tundra. Additionally, in the discontinuous permafrost zone, areas underlain by permafrost zones (i.e. SOC enriched zones) can be limited by topography in the subarctic . In Stordalen, this manifests as the lowest elevation areas being host to the peatland mire.
Of all the variables used in the correlation matrix, the strongest significant positive relationship with DOC was the mire class. Additionally, the mire class was consistently positive in all our LMER models. However, mires also had significant positive correlations with lakes and the NDVI. The influence of mires begins to explain the modest correlation value between the NDVI and DOC. Even though mires have low NDVI values, their role in subsurface contributions of DOC cannot be ignored.
In contrast to mires, lakes in all LMER models were found to have a strong negative effect on DOC. This observation suggests that likely no major autochthonous production of DOC occurs in these lakes, but rather lakes serve as transformation zones where DOC is lost. In Stordalen, lakes had a significant positive correlation with mires. This arises due to spatial arrangement in the catchment i.e. lakes and mires are often adjacent to each other or within the same elevation zone. Conversely, tundra and slope had significantly negative relationships with lakes.
As expected, birch forest had its highest positive and significant correlation with the NDVI followed by DOC. In the LMER models, the birch forest had an antagonistic effect on DOC. Generally, the proportions of mire and forest classes are used to predict DOC export in the boreal region . However, the forest class in this case was negative. Although a primary predictor of DOC in Stordalen, the negative sign of forest in all the LMER models suggests that we assess the robustness of the NDVI as a proxy, particularly weaknesses in the manner in which it deals with saturation . The NDVI has been shown to perform weakly in dense conifer forests in the boreal region . These factors compounded by our assessment period (May–Sept), begin to explain the negative sign of the forest land cover class, even though it is the largest class in the catchment.
Significant negative relationships with DOC were found with topographic indicators slope and tundra. Accordingly, in Stordalen higher DOC can be expected where there are mire and forest land cover classes. Periphery subcatchments associated with tundra and rock outcrop classes (e.g. A7, B2) had low NDVI values in Fig 3. Similarly, maps of DOC distribution show that the upper zones of periphery subcatchments contributed least to DOC.
The potential role of groundwater cannot be overlooked in influencing DOC concentrations in Stordalen and our findings. Subcatchments that receive regional groundwater inputs (e.g. B1, B2) had lower DOC concentrations as a result of higher runoff and electrical conductivity . Groundwater, a variable that causes variability in DOC concentration, is not accounted for in our models. Hence, our DOC predictions are limited in this regard.
The DOCqwc was incorporated to give a broad picture of DOC dynamics from the catchment. In some cases such as slope and NDVI, the DOCqwc provided estimates of long term DOC interactions. For elevation DOCqwc did not provide meaningful results in our analysis. However, what became evident with DOCqwc results was their dependency on dilution. In the future, a longer flow record and scaling may show better spatio-temporal differences, which could benefit our DOCqwc analysis.
5. Summary and conclusion
As inland waters are becoming more important in global carbon studies, it is important to understand the role that the surrounding landscape contributes to the measured DOC. Rivers are dynamic systems that, in contrast to lakes, flow through different land cover classes and media. They receive inputs from subsurface and overland (infiltration and saturation excess) flow, as well as precipitation, which change sometimes at fast rates. Climate change predictions show increases in vegetation. High terrestrial productivity is largely associated with increased DOC export. In Stordalen catchment we tested this hypothesis with terrestrial productivity approximated by the NDVI. The hypothesis holds some truth, however; it (NDVI) is not a strong indicator of DOC export on its own. Using simple regression methods this becomes clear. Stringent statistical tests using linear mixed regression modeling support the use of land cover as a primary indicator of DOC export in a catchment. Variables such as NDVI, elevation, and slope should be used in conjunction with land cover data (e.g. mires, forests, lakes). In particular, the role and influence of mires is highlighted. In this study we found that when terrestrial productivity increases, the DOC concentration in recipient waters also slightly increases. However, if peatlands are affected by increased precipitation, altered hydrology, and frost conditions etc., that might have a larger bearing on the lateral transfer of DOC. However, more research is needed in other northern catchments to reach a conceptual understanding of the controls influencing DOC release which are strongly affected by climate change. This does not have to be restricted to northern catchments but research from other geographic areas with differing spatio-temporal configurations can enrich the narrative of the terrestrial lateral transfer of carbon.
S1 Text. Source data citation aprroval for the vegetation map of Stordalen catchment, Sweden.
S2 Text. ORNL DAAC graphics citation communication.
S1 Protocol. Guide to attribution of geodata by Lantmäteriet.
We acknowledge the EUROSA Project (Erasmus Mundus Action 2 Partnerships scholarship programme) for facilitating the author’s exchange at Lund University, Sweden. Many thanks to Per Schubert who sat in early discussions for this work. The anonymous reviewers are thanked for their comments which have improved this paper.
- 1. Schuur EAG, Vogel JG, Crummer KG, Lee H, Sickman JO, Osterkamp TE. The effect of permafrost thaw on old carbon release and net carbon exchange from tundra. Nature. 2009;459: 556–559. pmid:19478781
Zhang T, Barry RG, Knowles K, Ling F, Armstrong RL. Distribution of seasonally and perennially frozen ground in the Northern Hemisphere. Proceedings of the 8th International Conference on Permafrost. AA Balkema Publishers; 2003. pp. 1289–1294.
Ciais P, Sabine C, Bala G, Bopp L, Brovkin V, Canadell J, et al. Carbon and other biogeochemical cycles. Climate change 2013: the physical science basis Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press; 2014. pp. 465–570.
- 4. Zhang X, Zwiers FW, Hegerl GC, Lambert FH, Gillett NP, Solomon S, et al. Detection of human influence on twentieth-century precipitation trends. Nature. 2007;448: 461–465. pmid:17646832
- 5. Kothawala DN, Stedmon CA, Müller RA, Weyhenmeyer GA, Köhler SJ, Tranvik LJ. Controls of dissolved organic matter quality: evidence from a large-scale boreal lake survey. Glob Change Biol. 2014;20: 1101–1114. pmid:24343949
- 6. Vonk JE, Tank SE, Bowden WB, Laurion I, Vincent WF, Alekseychik P, et al. Reviews and syntheses: Effects of permafrost thaw on Arctic aquatic ecosystems. Biogeosciences. 2015;12: 7129–7167.
- 7. Walvoord MA, Striegl RG. Increased groundwater to stream discharge from permafrost thawing in the Yukon River basin: Potential impacts on lateral export of carbon and nitrogen. Geophys Res Lett. 2007;34: L12402.
- 8. Borge AF, Westermann S, Solheim I, Etzelmüller B. Strong degradation of palsas and peat plateaus in northern Norway during the last 60 years. The Cryosphere Discuss. 2016;2016: 1–31.
- 9. Hallinger M, Manthey M, Wilmking M. Establishing a missing link: warm summers and winter snow cover promote shrub expansion into alpine tundra in Scandinavia. New Phytologist. 2010;186: 890–899. pmid:20345642
- 10. Giesler R, Lyon SW, Mörth C-M, Karlsson J, Karlsson EM, Jantze EJ, et al. Catchment-scale dissolved carbon concentrations and export estimates across six subarctic streams in northern Sweden. Biogeosciences. 2014;11: 525–537.
- 11. Hugelius G, Virtanen T, Kaverin D, Pastukhov A, Rivkin F, Marchenko S, et al. High-resolution mapping of ecosystem carbon storage and potential effects of permafrost thaw in periglacial terrain, European Russian Arctic. J Geophys Res. 2011;116: G03024.
- 12. Hodgkins SB, Tfaily MM, McCalley CK, Logan TA, Crill PM, Saleska SR, et al. Changes in peat chemistry associated with permafrost thaw increase greenhouse gas production. PNAS. 2014;111: 5819–5824. pmid:24711402
- 13. Swindles GT, Morris PJ, Mullan D, Watson EJ, Turner TE, Roland TP, et al. The long-term fate of permafrost peatlands under rapid climate warming. Sci Rep. 2015;5. pmid:26647837
- 14. Kirtman B, Power SB, Adedoyin JA, Boer GJ, Bojariu R, Camilloni I, et al. Near-term climate change: projections and predictability. Climate change. 2013; 953–1028.
- 15. Andersson A, Meier HEM, Ripszam M, Rowe O, Wikner J, Haglund P, et al. Projected future climate change and Baltic Sea ecosystem management. AMBIO. 2015;44: 345–356. pmid:26022318
- 16. Smith B, Aasa A, Ahas R, Blenckner T, Callaghan TV, Chazal J de, et al. Climate-related Change in Terrestrial and Freshwater Ecosystems. Assessment of Climate Change for the Baltic Sea Basin. Springer Berlin Heidelberg; 2008. pp. 221–308.
- 17. Hedenås H, Olsson H, Jonasson C, Bergstedt J, Dahlberg U, Callaghan TV. Changes in Tree Growth, Biomass and Vegetation Over a 13-Year Period in the Swedish Sub-Arctic. AMBIO. 2011;40: 672–682. pmid:21954729
- 18. Monteith DT, Stoddard JL, Evans CD, de Wit HA, Forsius M, Høgåsen T, et al. Dissolved organic carbon trends resulting from changes in atmospheric deposition chemistry. Nature. 2007;450: 537–540. pmid:18033294
- 19. Kritzberg ES, Ekström SM. Increasing iron concentrations in surface waters–a factor behind brownification? Biogeosciences. 2012;9: 1465–1478.
- 20. Brothers S, Köhler J, Attermeyer K, Grossart HP, Mehner T, Meyer N, et al. A feedback loop links brownification and anoxia in a temperate, shallow lake. Limnol Oceanogr. 2014;59: 1388–1398.
- 21. Jonsson M, Hedström P, Stenroth K, Hotchkiss ER, Vasconcelos FR, Karlsson J, et al. Climate change modifies the size structure of assemblages of emerging aquatic insects. Freshw Biol. 2015;60: 78–88.
Graneli W. Brownification of lakes. Encyclopedia of lakes and reservoirs. Springer; 2012. pp. 117–119.
- 23. Mormul RP, Ahlgren J, Ekvall MK, Hansson L-A, Brönmark C. Water brownification may increase the invasibility of a submerged non-native macrophyte. Biol Invasions. 2012;14: 2091–2099.
- 24. Lennon JT, Hamilton SK, Muscarella ME, Grandy AS, Wickings K, Jones SE. A Source of Terrestrial Organic Carbon to Investigate the Browning of Aquatic Ecosystems. PLOS ONE. 2013;8: e75771. pmid:24124511
- 25. Tuvendal M, Elmqvist T. Ecosystem Services Linking Social and Ecological Systems: River Brownification and the Response of Downstream Stakeholders. Ecology and Society. 2011;16.
- 26. Kritzberg ES, Granéli W, Björk J, Brönmark C, Hallgren P, Nicolle A, et al. Warming and browning of lakes: consequences for pelagic carbon metabolism and sediment delivery. Freshw Biol. 2014;59: 325–336.
- 27. Finstad AG, Andersen T, Larsen S, Tominaga K, Blumentrath S, Wit HA de, et al. From greening to browning: Catchment vegetation development and reduced S-deposition promote organic carbon load on decadal time scales in Nordic lakes. Scientific Reports. 2016;6: 31944. pmid:27554453
- 28. de Wit HA, Valinia S, Weyhenmeyer GA, Futter MN, Kortelainen P, Austnes K, et al. Current Browning of Surface Waters Will Be Further Promoted by Wetter Climate. Environ Sci Technol Lett. 2016;3: 430–435.
- 29. Roulet N, Moore TR. Environmental chemistry: Browning the waters. Nature. 2006;444: 283–284. pmid:17108948
- 30. Karlsson J, Byström P, Ask J, Ask P, Persson L, Jansson M. Light limitation of nutrient-poor lake ecosystems. Nature. 2009;460: 506–509. pmid:19626113
Jennings E, Järvinen M, Allott N, Arvola L, Moore K, Naden P, et al. Impacts of climate on the flux of dissolved organic carbon from catchments. The impact of climate change on european lakes. Springer; 2009. pp. 199–220.
- 32. Lapierre J-F, Guillemette F, Berggren M, Giorgio PA del. Increases in terrestrially derived carbon stimulate organic carbon processing and CO2 emissions in boreal aquatic ecosystems. Nature Communications. 2013;4: ncomms3972. pmid:24336188
- 33. Wit HA de, Ledesma JLJ, Futter MN. Aquatic DOC export from subarctic Atlantic blanket bog in Norway is controlled by seasalt deposition, temperature and precipitation. Biogeochemistry. 2016;127: 305–321.
- 34. Bäckstrand K, Crill PM, Jackowicz-Korczyñski M, Mastepanov M, Christensen TR, Bastviken D. Annual carbon gas budget for a subarctic peatland, Northern Sweden. Biogeosciences. 2010;7: 95–108.
- 35. Koven CD, Ringeval B, Friedlingstein P, Ciais P, Cadule P, Khvorostyanov D, et al. Permafrost carbon-climate feedbacks accelerate global warming. PNAS. 2011;108: 14769–14774. pmid:21852573
- 36. Raymond PA, Saiers JE, Sobczak WV. Hydrological and biogeochemical controls on watershed dissolved organic matter transport: pulse-shunt concept. Ecol. 2016;97: 5–16.
- 37. Aufdenkampe AK, Mayorga E, Raymond PA, Melack JM, Doney SC, Alin SR, et al. Riverine coupling of biogeochemical cycles between land, oceans, and atmosphere. Frontiers in Ecology and the Environment. 2011;9: 53–60.
- 38. Tang J, Miller PA, Persson A, Olefeldt D, Pilesjö P, Heliasz M, et al. Carbon budget estimation of a subarctic catchment using a dynamic ecosystem model at high spatial resolution. Biogeosciences. 2015;12: 2791–2808.
- 39. Ågren AM, Buffam I, Cooper DM, Tiwari T, Evans CD, Laudon H. Can the heterogeneity in stream dissolved organic carbon be explained by contributing landscape elements? Biogeosciences. 2014;11: 1199–1213.
- 40. Larsen S, Andersen T, Hessen DO. Climate change predicted to cause severe increase of organic carbon in lakes. Global Change Biology. 2011;17: 1186–1192.
- 41. Larsen S, Andersen T, Hessen DO. Predicting organic carbon in lakes from climate drivers and catchment properties. Global Biogeochem Cycles. 2011;25: GB3007.
- 42. Camino-Serrano M, Gielen B, Luyssaert S, Ciais P, Vicca S, Guenet B, et al. Linking variability in soil solution dissolved organic carbon to climate, soil type, and vegetation type. Global Biogeochem Cycles. 2014;28: 2013GB004726.
- 43. Olefeldt D, Roulet NT. Effects of permafrost and hydrology on the composition and transport of dissolved organic carbon in a subarctic peatland complex. J Geophys Res. 2012;117: G01005.
- 44. Lundin EJ, Giesler R, Persson A, Thompson MS, Karlsson J. Integrating carbon emissions from lakes and streams in a subarctic catchment. J Geophys Res Biogeosci. 2013;118: 1200–1207.
- 45. Petrescu AMR, van Huissteden J, Jackowicz-Korczynski M, Yurova A, Christensen TR, Crill PM, et al. Modelling CH4 emissions from arctic wetlands: effects of hydrological parameterization. Biogeosciences. 2008;5: 111–121.
- 46. Kokfelt U, Reuss N, Struyf E, Sonesson M, Rundgren M, Skog G, et al. Wetland development, permafrost history and nutrient cycling inferred from late Holocene peat and lake sediment records in subarctic Sweden. J Paleolimnol. 2010;44: 327–342.
- 47. Johansson T, Malmer N, Crill PM, Friborg T, Åkerman JH, Mastepanov M, et al. Decadal vegetation changes in a northern peatland, greenhouse gas fluxes and net radiative forcing. Global Change Biology. 2006;12: 2352–2369.
- 48. Kokfelt U, Rosén P, Schoning K, Christensen TR, Förster J, Karlsson J, et al. Ecosystem responses to increased precipitation and permafrost decay in subarctic Sweden inferred from peat and lake sediments. Global Change Biology. 2009;15: 1652–1663.
Lantmäteriet. GSD-Höjddata, grid 2+. In: GSD-Höjddata, grid 2+ [Internet]. 2016 [cited 24 Oct 2017]. Available: http://www.lantmateriet.se/sv/Kartor-och-geografisk-information/Hojddata/GSD-Hojddata-grid-2/
- 50. Deng J, Li C, Frolking S, Zhang Y, Bäckstrand K, Crill P. Assessing effects of permafrost thaw on C fluxes based on multiyear modeling across a permafrost thaw gradient at Stordalen, Sweden. Biogeosciences. 2014;11: 4753–4770.
Lal R, Stewart BA. Food Security and Soil Quality. CRC Press; 2010.
- 52. Wang Q, Ni J, Tenhunen J. Application of a geographically-weighted regression analysis to estimate net primary production of Chinese forest ecosystems. Global Ecology and Biogeography. 2005;14: 379–393.
- 53. Wessels KJ, Prince SD, Reshef I. Mapping land degradation by comparison of vegetation production to spatially derived estimates of potential production. Journal of Arid Environments. 2008;72: 1940–1949.
- 54. Rafique R, Zhao F, de Jong R, Zeng N, Asrar GR. Global and Regional Variability and Change in Terrestrial Ecosystems Net Primary Production and NDVI: A Model-Data Comparison. Remote Sensing. 2016;8: 177.
Yengoh GT, Dent D, Olsson L, Tengberg AE, Iii CJT. Key Issues in the Use of NDVI for Land Degradation Assessment. Use of the Normalized Difference Vegetation Index (NDVI) to Assess Land Degradation at Multiple Scales. Springer International Publishing; 2015. pp. 31–35. https://doi.org/10.1007/978-3-319-24112-8_5
- 56. Olefeldt D, Turetsky MR, Crill PM, McGuire AD. Environmental and physical controls on northern terrestrial methane emissions across permafrost zones. Glob Change Biol. 2013;19: 589–603. pmid:23504795
- 57. Hirsch RM, Moyer DL, Archfield SA. Weighted Regressions on Time, Discharge, and Season (WRTDS), with an Application to Chesapeake Bay River Inputs1. JAWRA Journal of the American Water Resources Association. 2010;46: 857–880. pmid:22457569
Maybeck M. Fluvial Export. In: Likens G.E., Biogeochemistry of Inland Waters. 1st ed. Amsterdam: Elsevier; 2009. p. 118–30.
ORNL DAAC. MODIS Collection 6 Land Products Global Subsetting and Visualization Tool. ORNL Distributed Active Archive Center; 2008. http://dx.doi.org/10.3334/ORNLDAAC/124111.
Ravindranath NH, Ostwald M. Carbon Inventory Methods: Handbook for Greenhouse Gas Inventory, Carbon Mitigation and Roundwood Production Projects. Springer Science & Business Media; 2007.
- 61. Ågren A, Buffam I, Jansson M, Laudon H. Importance of seasonality and small streams for the landscape regulation of dissolved organic carbon export. J Geophys Res. 2007;112: G03003.
- 62. Kicklighter DW, Hayes DJ, McClelland JW, Peterson BJ, McGuire AD, Melillo JM. Insights and issues with simulating terrestrial DOC loading of Arctic river networks. Ecological Applications. 2013;23: 1817–1836. pmid:24555311
- 63. Panneer Selvam B, Lapierre J-F, Guillemette F, Voigt C, Lamprecht RE, Biasi C, et al. Degradation potentials of dissolved organic carbon (DOC) from thawed permafrost peat. Sci Rep. 2017;7. pmid:28378792
- 64. Ågren A, Berggren M, Laudon H, Jansson M. Terrestrial export of highly bioavailable carbon from small boreal catchments in spring floods. Freshwater Biology. 2008;53: 964–972.
- 65. Lauerwald R, Hartmann J, Ludwig W, Moosdorf N. Assessing the nonconservative fluvial fluxes of dissolved organic carbon in North America. J Geophys Res. 2012;117: G01027.
- 66. Johansson M, Christensen TR, Akerman HJ, Callaghan TV. What Determines the Current Presence or Absence of Permafrost in the Torneträsk Region, a Sub-arctic Landscape in Northern Sweden? AMBIO: A Journal of the Human Environment. 2006;35: 190–197.
- 67. Schubert P, Eklundh L, Lund M, Nilsson M. Estimating northern peatland CO2 exchange from MODIS time series data. Remote Sensing of Environment. 2010;114: 1178–1189.
- 68. Jin H, Eklundh L. A physically based vegetation index for improved monitoring of plant phenology. Remote Sensing of Environment. 2014;152: 512–525.
- 69. Olefeldt D, Roulet N, Giesler R, Persson A. Total waterborne carbon export and DOC composition from ten nested subarctic peatland catchments—importance of peatland cover, groundwater influence, and inter-annual variability of precipitation patterns. Hydrol Process. 2013;27: 2280–2294.