Science of the Total Environment 520 (2015) 260–269

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Patterns and predictability in the intra-annual organic carbon variability across the boreal and hemiboreal landscape Julia K. Hytteborn a,⁎, Johan Temnerud b,c, Richard B. Alexander d, Elizabeth W. Boyer e, Martyn N. Futter b, Mats Fröberg b, Joel Dahné c, Kevin H. Bishop a,b a

Department of Earth Sciences, Uppsala University, Villavägen 16, 752 36 Uppsala, Sweden Department of Aquatic Sciences and Assessment, Swedish University of Agricultural Sciences, Box 7050, 750 07 Uppsala, Sweden Swedish Meteorological and Hydrological Institute, Research Department, 601 76 Norrköping, Sweden d U.S. Geological Survey, Reston, VA 20192, USA e Department of Ecosystem Science and Management, Pennsylvania State University, 117 Forest Resources Building, University Park, PA 16802, USA b c

H I G H L I G H T S • • • •

We investigated the intra-annual variability in TOC concentration in 215 watercourses. Discharge and seasonality controlled most of the intra-annual variability. Catchment characteristics partly explained the controls. Climate change is likely to affect the intra-annual TOC variability.

a r t i c l e

i n f o

Article history: Received 19 November 2014 Received in revised form 9 March 2015 Accepted 9 March 2015 Available online 26 March 2015 Editor: D. Barcelo Keywords: Total organic carbon Concentration and load estimation Watercourses Seasonality Climate change Water quality modeling

a b s t r a c t Factors affecting total organic carbon (TOC) concentrations in 215 watercourses across Sweden were investigated using parameter parsimonious regression approaches to explain spatial and temporal variabilities of the TOC water quality responses. We systematically quantified the effects of discharge, seasonality, and long-term trend as factors controlling intra-annual (among year) and inter-annual (within year) variabilities of TOC by evaluating the spatial variability in model coefficients and catchment characteristics (e.g. land cover, retention time, soil type). Catchment area (0.18–47,000 km2) and land cover types (forests, agriculture and alpine terrain) are typical for the boreal and hemiboreal zones across Fennoscandia. Watercourses had at least 6 years of monthly water quality observations between 1990 and 2010. Statistically significant models (p b 0.05) describing variation of TOC in streamflow were identified in 209 of 215 watercourses with a mean Nash-Sutcliffe efficiency index of 0.44. Increasing long-term trends were observed in 149 (70%) of the watercourses, and intra-annual variation in TOC far exceeded inter-annual variation. The average influences of the discharge and seasonality terms on intra-annual variations in daily TOC concentration were 1.4 and 1.3 mg l−1 (13 and 12% of the mean annual TOC), respectively. The average increase in TOC was 0.17 mg l−1 year−1 (1.6% year−1). Multivariate regression with over 90 different catchment characteristics explained 21% of the spatial variation in the linear trend coefficient, less than 20% of the variation in the discharge coefficient and 73% of the spatial variation in mean TOC. Specific discharge, water residence time, the variance of daily precipitation, and lake area, explained 45% of the spatial variation in the amplitude of the TOC seasonality. Because the main drivers of temporal variability in TOC are seasonality and discharge, first-order estimates of the influences of climatic variability and change on TOC concentration should be predictable if the studied catchments continue to respond similarly. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Factors controlling the total organic carbon (TOC) concentrations in watercourses are of great interest since TOC is a critical water quality ⁎ Corresponding author. E-mail address: [email protected] (J.K. Hytteborn).

http://dx.doi.org/10.1016/j.scitotenv.2015.03.041 0048-9697/© 2015 Elsevier B.V. All rights reserved.

characteristic that regulates primary production and living conditions for aquatic biota (Karlsson et al., 2009; Kullberg et al., 1993), and affects treatability of drinking water (Eikebrokk et al., 2004; Ledesma et al., 2012; Zeng and Arnold, 2013). Increasing long-term trends in annual mean TOC concentrations in watercourses across the northern hemisphere have received considerable attention (Monteith et al., 2007; Clark et al., 2010; Monteith et al., 2014). In addition to widespread

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

changes in annual TOC concentration, there are also large intra-annual TOC variations (Winterdahl et al., 2014) that often exceed the year to year changes and yet have not been well studied. These rapid and large intra-annual (within year) changes are as important to aquatic life and drinking water treatability as long-term inter-annual (among years) changes. Questions remain about how patterns of intra-annual variation will respond to climate change. Drivers of inter-annual trends in TOC include recovery from acidification due to decrease in anthropogenic sulfate (SO24 −) deposition (Monteith et al., 2007; de Wit et al., 2007), changes in temperature (Freeman et al., 2001; Sarkkola et al., 2009), changes in precipitation and discharge (Köhler et al., 2008; Ågren et al., 2008) as well as combinations of these (Futter et al., 2009; Erlandsson et al., 2008). Considering the factors influencing short-term variability on the intra-annual scale, discharge and temperature are often discussed (e.g. Dawson et al., 2008; Köhler et al., 2009). Sulfate concentration, which can be influenced by drought (Clark et al., 2006), seasonal climate anomalies (Lepistö et al., 2014) and the temperature of preceding seasons (Ågren et al., 2010), have also been identified as factors affecting TOC. Changes in short-term variation due to climate change may alter ecosystem functioning both through direct effect such as discharge and temperature, or changes in ecosystem response to climatic drivers. That a limited number of hydro-climatic factors control a large amount of annual TOC variability provides a good basis for parameter parsimonious simulation of TOC at both intra-annual and inter-annual time scales. Different types of models have been used to estimate the TOC concentration in watercourses from parameter parsimonious models (e.g. Grieve, 1991; Boyer et al., 2000) to complex processbased models (Yurova et al., 2008; de Wit et al., 2007; Futter et al., 2007; Ledesma et al., 2012). The complex models aspire to capture the intricacies of how present and changing climate can influence catchment functioning. There is, however, a dearth of data against which to evaluate the hypothesized process interactions in such models and the associated parameters. Parameter parsimonious statistical models can be calibrated more systematically than more complex conceptual or process-based models, but they predict on the assumption that the ecosystem will behave in the future as it has in the past. Simpler approaches such as this can be useful for detecting if a system is changing over time (Gebrehiwot et al., 2013). The aim of this study is to quantify the major influences on inter- and intra-annual TOC concentrations over decadal time scales across a large region using a standard load estimation modeling framework. This approach provides a unique opportunity to evaluate hypotheses about how catchment characteristics affect the intra-annual variability of TOC by using the spatial patterns in the model parameterizations. To estimate TOC concentrations in watercourses over large spatial and long temporal time scales requires long-term monitoring data of TOC concentration and discharge measured in numerous watercourses throughout a region, as well as their catchment characteristics. TOC monitoring data from 215 watercourses was used to quantify intraannual variability and trends in TOC. The watercourses are distributed across Sweden with catchments of different characteristics and sizes. The catchment data represent the primary land cover types found in the boreal and hemiboreal zone in all of Fennoscandia, including forested, agricultural, and alpine terrain. Parameter parsimonious regression models were applied to the TOC concentration data from all 215 watercourses, to evaluate how much of the intra- and inter-annual variations in TOC loadings (or export) could be explained by discharge, seasonality and long-term trend. The extent to which catchment characteristics can explain the spatial variation of the coefficients in the regression models was also quantified to gain understanding about what affects the TOC intra- and inter-annual variabilities. Whether the model residuals are stable over time was further evaluated to determine if catchment sensitivity to climate variables may be changing. This work provides a systematic assessment of the influence of discharge, seasonality and long-term trend on intra- and inter-annual

261

variations in TOC concentration of watercourses spanning a broad range of catchment size, land cover, latitude and climate. The application of a simple and consistent model specification to all watercourses is an important conceptual design element in this study and provides a uniform statistical approach for first quantifying intra- and interannual TOC variabilities and then assessing the major hydro-climatic and biophysical drivers of this variability across a diverse collection of catchments. 2. Materials and methods 2.1. Study area and model data sources The TOC data and other water chemistry variables for watercourses throughout Sweden were obtained from the database of the Swedish national environmental monitoring program administered by the Department of Aquatic Sciences and Assessment at the Swedish University of Agricultural Sciences, SLU (Fölster et al., 2014). The sampling was conducted in accordance with international standard SS-EN ISO 5667–1:2007, edition 1. In large watercourses, every attempt was made to collect water from the middle of the stream by sampling from a bridge using a Ruttner water sampler. If this was not possible, samples were collected from the shore. In small watercourses, water was collected by a grab sample with the sampling bottle. All available sample data from watercourses having at least 6 years of monthly TOC data within the 21-year period from 1990 to 2010 were deemed sufficient for use in the study. TOC concentration and discharge data were available for the concentration estimation modeling from 215 watercourses across Sweden, including more than 42,500 water quality samples where TOC was observed. The study area covered most of the area of Sweden, spanning a latitudinal gradient from 55° to 68° N (Fig. 1a). The median length of time the data sets was 15 years, and 90% of the watercourses had 10 or more years of data. The watercourses had a range in mean TOC concentration from 1.4–25 mg l−1 (Fig. 1b). Catchments had a range in average air temperature from − 2.0 to 8.8 °C (Fig. 1g); and a range in precipitation from 440 to 1270 mm per year (Fig. 1i). Temporal variation is reported as the coefficient of variance (CV, the standard deviation divided by the mean value) for TOC (Fig. 1c) and precipitation (Fig. 1j), and standard deviation (SD) is used for air temperature (Fig. 1h). The catchment areas range from 0.18 km2 to over 47,000 km2 (Fig. 1d). In order to keep observations independent, no nested subcatchments larger than a few percent of a downstream catchment area were included in the data set. The TOC data were all analyzed at the national laboratory at the Department of Aquatic Sciences and Assessment using a Shimadzu TOC-VPCH analyzer. The dissolved organic carbon (DOC) concentration is not analyzed. In Swedish watercourses it has previously been shown that DOC and TOC generally differ by less than a few percent (Ivarsson and Jansson, 1994; Köhler, 1999; Laudon et al., 2011). Time series plots of the TOC data from all 215 watercourses were manually inspected and a total of 9 TOC concentration outliers were removed. These outliers were all isolated, exceptionally high values at least twice as high as any other value in the time series. Daily discharge data (Fig. 1e and f) used in the TOC concentrationestimation models was computed by the Swedish Meteorological and Hydrological Institute (SMHI) using the Hydrological Predictions for the Environment (HYPE) model (Lindström et al., 2010; Strömqvist et al., 2012). When daily discharge modeled with HYPE was compared with daily observed values, the Nash-Sutcliffe efficiency (NSE, see Supplementary information S3) in the median watercourse was 0.68 and for long-term means the NSE was 0.997 for discharge and 0.92 for specific discharge with a median absolute error of 8% for both (Strömqvist et al., 2012). Time series of observed discharge between 1990 and 2010, gauged by SMHI was available at 59 of the 215 watercourses. These daily observed discharge data were used

262

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

a)

10°E

c)

b)

20°E

65°N

TOC, mean 1.4 - 7.6 7.6 - 10.6 10.6 - 14.2 14.2 - 24.5

60°N

TOC, CV 0.11 - 0.22 0.22 - 0.28 0.28 - 0.39 0.39 - 1.12

55°N

d)

e)

Area 0 - 40 41 - 500 501 - 47033

g)

f)

Specific discharge, mean 0.0041 - 0.0093 0.0093 - 0.012 0.012 - 0.015 0.015 - 0.031

h)

Temperatur, mean -2.0 - 3.3 3.3 - 6.4 6.4 - 7.1 7.1 - 8.8

i)

Temperature, SD 6.7 - 7.4 7.4 - 8.0 8.0 - 8.9 8.9 - 11.0

Specific discharge, CV 0.29 - 0.88 0.88 - 1.19 1.19 - 1.52 1.52 - 2.13

j)

Precipitation, annual 440 - 661 661 - 723 723 - 830 830 - 1266

Precipitation, CV 1.52 - 1.91 1.92 - 2.02 2.03 - 2.10 2.11 - 2.35

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

263

Table 1 Data used in the study. Type of data

Description

Reference

TOC concentration data Area and elevation

Water quality data base Catchment area and the mean elevation of the catchment area computed from the Swedish national 50 m grid data Theoretical water residence time calculated from estimation of lake volume from map parameters Daily data, modeled with S-HYPE at SMHI and corrected with upstream observed data Gauged data from SMHIs data base Daily data from the PTHBV database, 4-km grid data based on stations CORINE 2000 Soil type derived from database with scales 1:50,000, 1:100,000, 1:200,000 and 1:1.1000000, the best resolution thereof for each catchment. kNN database from Department of Forest Resource Management, SLU Modeled data based on elevation, coordinates and topographic wetness index

Fölster et al. (2014) The Swedish National Land Survey

Retention time

Discharge data, modeled Discharge data, observed Temperature and precipitation data Land use Soil type

Forest geographical data Carbon and nitrite content in soil

to evaluate whether differences between modeled and observed daily discharge influenced the prediction of TOC. References to all data can be found in Table 1. To evaluate potential causes of geospatial variability in the regression models describing intra- and inter-annual variabilities in TOC concentrations in the 215 watercourses, additional variables characterizing land cover, soil, physiographic and other catchment properties were calculated from national geospatial data sets. Catchment area and mean elevation were taken from the Swedish National Land Survey 50 meter grid data. Daily air temperature and precipitation were obtained from a database with spatially interpolated air temperature and precipitation on a 4 km grid, from meteorological station data, the database is called PTHBV (Johansson, 2002). For this study the daily spatial mean from the grid points in each catchment was computed. Land cover data were obtained from the European Environment Agency CORINE database (Bossard et al., 2000) and soil type data from the Geological Survey of Sweden (SGU) soil map. Forest data came from the SLU Forest Map (Reese et al., 2003) which reports the mean age of forest stands, and volume of forest biomass. Both wetlands and organic soils in forest areas are documented sources of TOC in watercourses (Laudon et al., 2011). A list of the over 90 different catchment characteristics used in the study can be found in the Supplementary information (S1, Table S1) in addition to basic statistics for TOC, discharge, precipitation, air temperature and several catchment characteristics summarized over relevant classes (Table S3). All water chemistry, discharge, precipitation, temperature and catchment characteristics from the 215 watercourses are publicly available from http://www.slu.se/cleo/data. 2.2. Parameter parsimonious modeling and model performance To facilitate the systematic application of TOC concentration estimation models, the Fluxmaster software package (Schwarz et al., 2006) developed by the U.S. Geological Survey was used. Fluxmaster implements a common and accepted approach to load estimation that provides a statistical optimization framework for estimating long-term concentrations and loadings of solutes from monitoring data. Using Fluxmaster, it is possible to estimate concentration and loads for multiple water quality constituents and watercourses, and is similar to other

Computed using the method of Müller et al. (2013) and Sobek et al. (2011) Lindström et al. (2010), Strömqvist et al. (2012)

Johansson (2002) Bossard et al. (2000) SGU (2012)

SLU Forest Map, Dept. of Forest Resource Management, Swedish University of Agricultural Sciences,Reese et al. (2003) (Supplementary information, S1)

popular tools for estimating constituent loads in rivers (Runkel et al., 2004). During the calibration stage, regression models were developed for the estimation of TOC concentration from time series of daily discharge and monthly TOC concentrations. The model estimation and predictions account for the effects of retransformation bias. Models were calibrated and evaluated using daily discharge and TOC data from one day each month. The calibration method was ordinary least squares or, if there are censored values or serial correlation in the data, maximum likelihood estimation. During the estimation stage, the regression models were used to quantify daily TOC concentrations throughout the 1990–2010 time period for each watercourse. To systematically compare patterns of TOC in different watercourses, the same regression model of daily TOC concentration was applied within Fluxmaster to describe the TOC temporal variability observed in all of the 215 watercourses. Model coefficients were calibrated separately for each watercourse (Eq. (1)). The model included discharge, a sine term for seasonality and long-term trend.

a0

TOC ¼ e

a1

 Discharge

A sinð2πdtimeþcÞ

e

a4 dtime

e

1

TOC is the modeled TOC concentration, a0, a1, A, c and a4 are model coefficients, Discharge is the daily discharge and dtime is decimal time. The four terms in the model are hereafter referred to as: the constant (ea0 ), discharge (Dischargea1 ), seasonality (eA ⋅ sin(2π ⋅ dtime + c)), and trend (ea4 dtime ) terms where a1 is the discharge coefficient and a4 is the trend coefficient. The coefficients in the seasonality term are A, the amplitude and c, the displacement. All references to amplitude in the text refer to the A coefficient in the seasonality term. The day when the seasonality term has its highest value is the peak day calculated from coefficient c. See S2 in the Supplementary information for more details. To evaluate the performance of the regression model (Eq. (1)), several goodness of fit measures were computed for each watercourse, including the r2, the NSE and root mean square error (RMSE), see Supplementary information S3. Statistical significance (p b 0.05) of the regression model and the associated coefficients in Eq. (1) were also quantified. Model performance was further investigated with respect

Fig. 1. The 215 study catchments, a) study area, b) mean total organic carbon (TOC) concentration, mg l−1, c) CV for TOC concentration, d) distribution of different catchment sizes in the study area, km2 and Limes Norrlandicus the gray line e) mean specific discharge, m3 s−1 km−2, f) CV for specific discharge, g) mean air temperature, °C, h) standard deviation (SD) for the air temperature (°C), i) annual precipitation over the catchment, mm year−1, and j) CV in daily precipitation. The mean and CV-values are calculated with data from 1990 to 2010. The category limits in the subfigures, are the 25, 50 and 75th percentiles.

264

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

to mean absolute error, relative error as well as model misspecification and bias, as described in the Supplementary information S4. Several regression models were considered to predict daily TOC concentrations in Swedish watercourses. The model in Eq. (1) was chosen as the most appropriate method. An evaluation is provided in the Supplementary information (S4) of model misspecification and prediction bias associated with the Eq. (1) regression model in comparison with a more complex model with an additional squared discharge term. The goodness of fit was slightly higher for the model with a squared term but this was outweighed by the advantages of using the more parameter parsimonious model in Eq. (1). Models with fewer parameters than in Eq. (1) gave lower goodness of fit measures. The prediction bias for the model in Eq. (1) was found to be very small for most of the 215 watercourses; on average, predictions are within about 1% of the observed values at 50% of the sites and within 8% of the observed values at 90% of the sites, with a general tendency for the model to overpredict concentrations. The use of modeled discharge data was also tested. The TOC concentrations were modeled with both gauged and modeled, non-corrected discharge data in the 59 watercourses where both data series existed. To check whether catchment sensitivity to factors hypothesized to affect TOC concentrations was changing over time, residuals from the models developed for each watercourse (Eq. (1)) were split in two time periods having equal amounts of data in each. Basic statistics of the residuals were computed for the two periods, and the temporal variances were quantified with the Brown–Forsyth test (Brown and Forsythe, 1974). To explore how the different terms in the model influence the intraannual variability of predicted TOC concentration in each watercourse, the daily influences from the discharge, seasonality and trend terms in Eq. (1), were explored individually. Since the models are multiplicative, each entire model has to be included when computing the influence from each model term. To facilitate this analysis, each model with coefficients calibrated for a specific watercourse was used, but the input data in the calculation of the TOC concentration were changed. In this altered calculation the term of interest was set to a constant value. The influence of the term was then defined as the mean absolute difference between the TOC concentration from the model used in the normal way and the altered calculation with the term of interest set to a constant. To quantify the influence of the discharge term, discharge in each altered calculation was set to the mean discharge for the watercourse. To quantify the influence of seasonality, the seasonality term was set to a value of 1. Finally, to consider the influence of the nonlinear trend terms, the mean change over the whole period (1990–2010) was computed. The change was computed in an altered calculation where only the trend term was permitted to change and the other terms were set to constant values; the discharge values were set to the mean discharge for that watercourse and seasonality term was set to a value of 1. The concentration difference between the end and start of the period was defined as the influence of the trend term. Many studies have shown that rates of acidic atmospheric deposition have decreased in recent decades across Europe, as indicated by temporal trends in oxidized sulfur deposition and by declines of SO2− 4 concentrations in watercourses suggesting recovery from acidification (Monteith et al., 2007). Here, the load estimation modeling framework was used as a first approximation to consider if recovery from acidification can explain recent changes in the TOC concentrations observed in Swedish surface waters. Fluxmaster models for SO2− concentrations 4 in watercourses using Eq. (1) as described above for TOC, were developed so that relationships between the trends in TOC and SO2− could 4 be evaluated. As both air temperature data and the modeled seasonality term exhibit strong seasonal structure in terms of both means and variances, the correlation between these terms for each watercourse was assessed. For this analysis, linear regressions were performed between simulated daily values of the seasonality term and the daily air temperature of the catchment for each watercourse. A total of 365 regressions were

performed for each watercourse where the time series data were shifted in relation to each other by one day for each regression. The shift with the highest correlation was identified and defined as the lag time between the seasonality term and daily air temperature. 2.3. Spatial variability The topographical, climatologic and vegetative geographic boundary zone Limes Norrlandicus in Sweden (Fransson, 1965, Fig. 1d) was used to divide Sweden into two different geographical areas. In north of Limes Norrlandicus, the vegetation is boreal or alpine while to the south the vegetation is hemiboreal. The mountainous, high-elevation northwest watercourses are characterized by low TOC concentrations and a large spring flood resulting in large intra-annual coefficient of variance (CV) for specific discharge (Fig. 1b and f). The watercourses draining the 215 catchments are fairly evenly distributed with 94 north and 121 south of Limes Norrlandicus (Fig. 1d). Four different catchment size classes were defined with cutoff points at 24 km2, 66.4 km2 and 820 km2 (Table S12). Significant (p b 0.05) differences in model coefficients, TOC concentration and goodness of fit values between different groups of watercourses (north and south of Limes Norrlandicus and size classes) were investigated with two-way analysis of variance (ANOVA), with post-hoc analysis using Tukey's honest significant difference (HSD) test (Tukey, 1949). The catchments were also classified with respect to four dominant land cover types: agriculture, forest, wetland and mixed. One-way ANOVA was performed on these classes, with posthoc analysis using Tukey's HSD test. Since small areas of agricultural land have a large impact on water quality, 5% or more was the cutoff of classification as agriculture. For the forest land cover class, 80% was the cutoff and for the wetland class, the cutoff was 20%. All others catchments were classified as mixed land cover. Catchments that meet two of the criteria are also classified as mixed. Watercourses having models with significant discharge and seasonality coefficients were used in all subsequent analysis of factors influencing TOC. In the Supplementary information, S5 (Table S13), basic statistics from the different groups are provided. Partial least squares regression (PLS, Geladi and Kowalski, 1986) was used to explore if catchment characteristics (e.g., land cover, soils, forest biomass, precipitation, temperature and discharge) could explain spatial variability in the model coefficients (Eq. (1)). Further details about the PLS analysis are included in the Supplementary information (S5). 3. Results 3.1. Model performance Regressions using Eq. (1) were able to generate statistically significant (p b 0.05) models from the available TOC data in 209 of the 215 watercourses (see Supplementary information, S3, Table S5). The six watercourses with non-significant models all had catchments larger than 100 km2, located in the south of Sweden, with low intra-annual variability in TOC concentrations (CV-values below the median CV for all watercourses). The median NSE value for the significant 209 models was 0.44 and the median r2-value was 0.46 (Table S6, Fig. S1). All three model terms (discharge, seasonality, and trend) were statistically significant in 109 (51%), of the watercourses. In 149 watercourses (69%), both discharge and the seasonality terms were significant. The discharge term was significant in 85% of the watercourses; the seasonality term was significant in 78%; and the trend term was significant in 70%. For 3 watercourses (b 2%), high correlation between discharge and seasonality caused both variables to be statistically insignificant in the regression models; however, our analyses of TOC variability were not sensitive to these multi-collinearity effects because so few sites are affected. The average NSE value was similar regardless of whether observed or modeled discharge data were used with an NSE of 0.39 for observed discharge data versus NSE of 0.40 for HYPE modeled discharge data in

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

a

The terms

ln(TOC)

4

b

4

2

265

c

4

2

2 Modeled TOC Sampled TOC

4

4

2

2

2

0

0

0

2000

Intercept Discharge term

2002

2004

2006

2008

2010 2000

2002

2004

2006

2008

2010

4

2000

Seasonality term Trend term

2002

2004

2006

2008

2010

Fig. 2. The TOC concentration and model term in three watercourses: Verkån, Haväng (a), Björkeredsbäcken (b), and Blankan Ryerna (c). In the upper part of each subfigure the logarithm of the observed TOC concentration is illustrated by brown crosses and the logarithm of the modeled TOC concentration is illustrated as a black line. The model TOC concentration is the sum of intercept, discharge, seasonality and long-term trend terms which are shown separately in the lower parts of each subfigure. The model and all terms were significant in the three examples. Observe that the y-axes are logarithmic.

59 gauged watercourses, indicating that the use of simulated discharge values does not affect model results. Discharge, seasonality, and trend components of Eq. (1) generate different amounts of intra-annual variability in TOC concentrations, which is exemplified by three watercourses (Fig. 2). Modeled and observed TOC concentrations as well as the individual model terms are estimated in log space, which means that the terms are added to compute the logarithm of the modeled TOC concentration. In watercourse a, the discharge term had the greatest influence on TOC concentration variability and the trend increases the baseline TOC concentration over time. In watercourse b both the discharge and seasonal term influence variability in TOC concentrations, and in watercourse c the seasonality term completely dominates the TOC concentration variability. In both watercourses b and c, a seasonal pattern was visible in the TOC concentration and the longterm trends were small. The residuals during the first and second half of the study periods were investigated to determine if the model performance changed over time. Basic statistics are presented in the Supplementary information (S1, Table S4). For 96% of the watercourses there was no significant change in the variance of the residuals between the two time periods according to the Brown–Forsyth test. Of the eight watercourses where there were significant changes, five were located northwest of Lake Vänern, in an area that experienced severe flooding in 2000, which was at the breakpoint between the two time periods used for this analysis. After the flooding, several samples with very high TOC concentrations were reported in these five watercourses, which changed the variance in the residuals. Analyses of the specific influence of discharge and seasonality terms on intra-annual variability were based on the absolute change (in mg l−1) in TOC concentration. Influences were higher in watercourses with high TOC concentration. Considering all the watercourses, the daily mean influence on TOC concentration from discharge and seasonality terms was of the same magnitude (Table S6). The daily mean influence of seasonality in the 95th percentile watercourse was 4.2 mg l−1 compared to 3.3 mg l−1 for the discharge influence. The mean change due to the long-term trend was 0.17 mg l−1 year−1. The model coefficients are a measure of the relative impact of the different drivers on TOC concentration. The mean discharge coefficient and amplitude in the seasonality term were of similar magnitude, 0.15 and 0.18, respectively, whereas the trend was an order of magnitude lower, 0.016. The discharge coefficient and amplitude had spatial CV values of 0.83 and 0.74, which were higher than the spatial CV value of the mean observed TOC (0.44). This means that the factors influencing the intra-annual TOC variability are more spatially variable than the factors influencing mean TOC concentration.

The mean correlation between Fluxmaster modeled SO2− and TOC 4 concentrations was 0.59. In 138 watercourses the trend terms were significant for both the SO2− and the TOC models. The SO2− trend coeffi4 4 cients were negative in 96% of the watercourses, indicating a decrease in concentration over time. The linear regression between SO24 − and TOC trend coefficients was negative with a very low r2-value of 0.03 (Fig. S4). The seasonality term has, like the air temperature, a period of one year. The time lags between the time series in air temperature and the seasonality term had mean and median values of 71 and 51 days, respectively, for watercourses with significant seasonality terms. This means that the highest air temperature, which usually was in the beginning of July, was often followed by the peak in the seasonality term around 2 months later in many watercourses. The correlation between seasonality term and air temperature values was reasonably strong with r2-values between 0.66 and 0.81, all of which were significant at a p-value b 0.0001.

3.2. Spatial variability Spatially, the influence of the discharge and seasonality terms on modeled TOC was fairly heterogeneous (Fig. 3a, b and d) and does not co-vary with each other (Fig. S3). Visual scrutiny suggests some regional patterns in the influence of the different terms (Fig. 3). Both the seasonality and discharge terms had a high influence on TOC concentrations in the southern Swedish highlands (Fig. 3a and b). The discharge term had a few high values spread out in the north. The seasonality term had an east–west gradient in the north with lower influence in the west and higher values towards the east coast, Fig. 3b. In watercourses with significant models, 65% had a larger discharge than seasonality influence, but in six watercourses, all with an area less than 40 km2 located in southwest Sweden, the seasonality term completely dominated the influence on TOC variability (Fig. 3d). In the 95% elliptic confidence intervals, the distance from origin represents the size of the amplitude and the angle represents the peak day (Figs. 3f, g and S6a to h). For most of the watercourses with small catchments north of Limes Norrlandicus (N1 and N2), the peak day occurred in the third quarter of the year, whereas the largest catchment class (N4) had watercourses with large amplitudes and peak days in the first and third quarters of the year (Fig. 3f). For the watercourses south of Limes Norrlandicus, the smallest catchment class (S1) had the highest amplitudes with gradually smaller amplitudes in larger catchments (Figs. 3g and S6b, d, f and h). The peak day in the seasonality term was usually in the third quarter of the year in watercourses in all four catchment classes south of Limes Norrlandicus.

266

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

Fig. 3. Information about the model terms a) influence from the discharge term in mg l−1 days−1, b) influence from the seasonality term in mg l−1 days−1, c) the 10 year contribution from the trend term in mg l−1°10 years−1, d) difference in influence from seasonality and discharge terms, e) the peak month in the seasonality, f) and g) annual plots showing 95% elliptic confidence intervals of amplitude and peak day from watercourses with significant seasonality term, f) north of Limes Norrlandicus where N1 is the class with smallest catchment area and N4 have the largest size, g) the watercourses south of Limes Norrlandicus. The limits of the color ranges for subfigures a), b) and c) are according to their 1st, 2nd and 3rd quartiles, subfigure c) have an extra category for the negative values and watercourses with non-significant model terms are marked with black semicircles. In the Supplemantary information (S5) a more detailed figure of f) and g) is found (Fig. S6).

In a majority of watercourses (69%) the Fluxmaster model had a significant positive trend, while two watercourses in the north had significant negative trends (Fig. 3c). Along the southeast and south coast the trend had high to intermediate influence values. There was also a cluster of watercourses with high trend influence north of Lake Vänern, the largest Swedish lake. The high trend in this watercourses might be related to the extensive flooding in the area in 2000 (Lindström and Bergström, 2004). In the years following the flooding, more samples with exceptionally high TOC concentration were observed. The geographical distribution of model coefficients and influence (Fig. S2) was quite similar but there were more watercourses with high and intermediate coefficients in the north, especially for discharge coefficients.

In the two-way ANOVA analysis with Tukey's HSD post-hoc tests, the constant term was significantly higher in small southern catchments than for the largest size class both north and south of the Limes Norrlandicus. This is indicative of the relatively high TOC concentrations in small southern watercourses. The discharge coefficients were higher in the intermediate catchment size class (2) than the smallest (1). Amplitudes were generally higher in small southern watercourses than in large rivers north of the Limes Norrlandicus and the trend coefficients were significantly higher south of Limes Norrlandicus than to the north. Both the NSE and r2 statistics were higher for southern rivers. There were no clear patterns for the other goodness of fit measures. The time lags between air temperature and the seasonality term were complicated but in general, southern rivers had longer time lags than

4. Discussion 4.1. The conceptual approach

a

See Table S1 for variable names and descriptions.

[SpecDischargeIQR, SpecDischargeMean] Peat, ConiForWet, PineMean, SpruceMean, CSoil,Mean ConiForWet, (Peat) 27/24 77/73 54/51 27/24 70/68 40/39 RE TOC mean sampled TOC SD sampled

ns/ns 7/16 13/20

[SpecDischargeMin, (OthDenseVegArea)] (SpecDischargeMean) DeciFor, Clay, Biomass, (OthPorVegArea) [Pp75, PAnnual] Pp75, PAnnual C:NsoilMean ConiForWet, Peat, PineMean 20/18 47/45 21/17 30/29 29/28 33/31 61/57 20/18 44/42 19/16 30/29 27/27 31/29 51/49 Discharge coefficient Amplitude Trend coefficient NSE r2 RMSE MAE

ns/ns 3/5 2/1 ns/ns 2/1 3/3 10/17

Negative

267

rivers north of the Limes Norrlandicus. The overall conclusion that can be drawn from these results is that small, southern rivers have a different response to climate, flow and long-term trend than large rivers in the north. Between land cover types, the forest constant term was higher than that for mixed land cover, suggesting higher baseline TOC in forests. The trend terms were lower in wetland-dominated watercourses than in those dominated by the other three land cover types. The time lags were higher in agricultural watercourses than in those dominated by other land cover types. The spatial variability in model coefficients could be related to catchment characteristics. This possibility was evaluated using PLS, with variable degrees of explanation for different parameters. The mean values of the sampled TOC concentration could also be modeled using PLS on catchment characteristics (measure of the model fit, R2 = 0.77 and robustness, Q2 = 0.73) with many catchment characteristics contributing, for example, negative correlations to discharge measures and elevation while there were positive correlations with temperature and peat area. Only small to negligible spatial variation in the discharge coefficient of Eq. (1) could be explained (b 20% with only one significant principal component) using the PLS analysis (Table 2). Forty-seven percent of the spatial variation in the amplitude of the seasonality term in different watercourses was explained by catchment characteristics (Table 2). Of the 90 investigated explanatory variables, a majority of the explained variation in amplitudes could be attributed to four variables: mean specific discharge had a positive correlation, whereas the CV in daily precipitation, theoretical water retention time, and the percentage of lake surface cover in the catchment had negative correlations. For the trend coefficients, 21% of the spatial variation was explained by combinations of six different characteristics. The trend coefficient had a positive correlation with the land cover type “other poorly vegetated areas”, mean forest biomass, volume of deciduous trees and percentage of catchment with clay as soil type, as well as negative correlations with elevation and median precipitation.

OthDenseVegArea, FragRoke, C:NsoilSd, DischargeMin, SpecDischargeMin, OthPorVegArea, Talus, Dischargep25, DischargeMean, Area, DischargeMedian, Dischargep75, DischargeMax, Dischargeiqr, (SpecDischargeMedian, SpecDischargep25, Block) [–] Residence time, (PCV, Lake) Elevation, (PMedian) [(PCV), Residence time] PCV, (Residence time) Tp75, CorrLag ClaySilt, Lakes volume, Dischargesd, DischargeMax, OthDenseVegArea, DischargeMean, Dischargeiqr, Dischargep75, DischargeMedian, C:Nsoilsd, FragRock, Dischargep25, Talus, Area [CleFellFor, Peat] OthDenseVegArea, FragRock, C:NSoil,sd, AgeMean, Elevation, C:NSoil,Mean, Talus, OthPorVegArea, QMax, Qsd, QMin, QMean C:Nsoilsd, Lakes volume, Dischargesd, DischargeMax, DischargeMean, Area, Dischargep75, DischargeMedian, Dischargeirq, Dischargep25, (OthDenseVegArea) –

Positive

4/10 73/72 Constant

77/75

R2/Q2 R2/Q2

R2cum/Q2cum

Factors in the PLS analysis with VIP N 1 (in parentheses 0.9 b VIP b 1)a Cumulative PC2 (%) PC1 (%) Y-variable

Table 2 The percentage explained by the catchment characteristics (R2) and a measure of the model's robustness (Q2) for the PLS-analysis of the regression model coefficients. The 149 watercourses with significant seasonality and discharge term were used in the PLS-analysis. Non-significant is denoted as ns. All catchment characteristics in Table S1 were used initially in the PLS analysis. Factors in the PLS models with variable importance on the projection (VIP) scoring over 1 are presented; factors with 0.9 b VIP b 1 are shown in parentheses; factors in brackets belong to non-significant PLS models.

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

Based on literature studies, spatial and temporal variations in watercourse TOC (both intra- and inter-annual) are known to be controlled by a variety of hydro-climatic and biogeochemical factors, including runoff, temperature, land cover and sulfate. Many of these factors have been evaluated in simple and complex models to advance the understanding of how specific processes combine to influence watercourse TOC variability in different environmental settings. However, there have been few systematic efforts to quantify how the major influences compare to each other across broad spatial domains over the course of decades. Our approach addresses this need by modeling daily watercourse TOC concentrations during the past several decades using a set of long-term watercourse monitoring data that broadly represents the environmental conditions of catchments across Fennoscandia. We used a consistent, parameter parsimonious regression modeling technique to simulate intra- and inter-annual TOC variabilities as a function of discharge, seasonality and long-term trend. The models were calibrated with monitoring data from 215 catchments in Sweden between 1990 and 2010, where standard methods of sample collection and analysis have been employed. We then explored how the three major influences of discharge, seasonality and long-term trend on temporal variability are associated with a comprehensive set of watershed properties related to major hydro-climatic and biophysical factors. 4.2. Temporal variability The results demonstrate that a statistical model with discharge, seasonality and long-term trend terms can explain on average, 46%

268

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

(mean r2-value of 0.46) of the temporal variation in the TOC concentration data from 209 watercourses spread across a 1450-km-long region with differences in catchment size, land cover and climate. The load estimation model (Eq. (1)) was able to capture much of the observed inter- and intra-annual variability in TOC concentrations over time throughout the boreal and hemiboreal regions of Sweden. While more complex, process-oriented approaches can be used for modeling, the simplicity of this approach gave a readily interpretable and widely applicable outcome. The goodness of fit was comparable to what other models of TOC have been achieving on fewer, more intensively studied sites with many more parameters to calibrate (Futter and de Wit, 2008; Jutras et al., 2011). For many of the studied watercourses, the seasonality term had its peak in August or September. This can be an indication that the seasonality term was a proxy variable for the soil temperature or rather biological activity and decomposition, which has been shown to have a relation to TOC concentration along with discharge (Ågren et al., 2008; Köhler et al., 2008). Adding soil temperature to models where discharge was the only predictor variable has been shown to improve the prediction of TOC concentrations in other studies of boreal watercourses (Winterdahl et al., 2011). To use soil temperature as a possible improvement in the expression of the seasonality term would be an important factor to evaluate in future studies focused on assessing the effects of climate change on stream organic carbon.

4.3. Spatial variability Since the same model is applied consistently, it is possible to compare the spatial and temporal patterns in model terms. In this study, discharge and seasonality have a similar degree of influence, but there were some spatial differences. The influences of both the discharge and seasonality terms were in general largest in areas of Sweden where mean TOC concentrations were highest. This was in the southern Swedish highlands and along a northern section of the east coast. Mean TOC concentrations were also high in a region just north of Lake Mälaren on the east coast, but the influence of discharge and seasonality terms were low there. North of Limes Norrlandicus the discharge term had a larger impact on TOC concentrations in large catchments than in small catchments, while south of Limes Norrlandicus seasonality had a larger impact in small catchments than in large catchments. The overall patterns identified in this study are similar to those found by Winterdahl et al. (2014) in a statistical evaluation of TOC data from Swedish watercourses. Fitting of model coefficients to each catchment, however, facilitates investigation of the influence of specific drivers on intra-annual TOC variability in relation to specific catchment characteristics. The processes responsible for the spatial patterns in the model coefficients cannot be fully explained by catchment characteristics included in the PLS analysis but the results provide insights and clues for further studies. Key findings from this study are the complexity of the relationship between catchment characteristics and the drivers of catchment sensitivity to discharge, seasonality and long-term trends. The PLS analysis of the Fluxmaster model coefficients showed that the discharge coefficients were not significantly explained (b 20% by one significant principal component) by any combination of the 90 catchment characteristics that were explored, including climate statistics, land cover, soil type and forest biomass. The discharge coefficients north of Limes Norrlandicus were, however, significantly higher in large catchments than in small catchments. On the other hand, the PLS analysis was able to explain some 45% of the spatial variation in the amplitude of the seasonality using four catchment characteristics (Table 2). Catchments that are wet (high specific discharge), with low CV in daily precipitation, little or no lake surface coverage, and short theoretical landscape water retention times have larger amplitudes in seasonality. Consequently, catchments with long retention times and large lake surface coverage have smaller amplitudes, presumably because of mixing

of TOC from different seasons as well as in-lake processes, that attenuate the signal from the seasonality. Only 20% of the spatial variation in trend coefficients could be explained by catchment characteristics in the PLS analysis. Two leading theories that explain the recent trend in TOC concentration include recovery from acidification and changes in climate. Both Monteith et al. (2007) using time series data from 522 lakes and watercourses and Futter et al. (2009) came to the conclusion that most of the reduction from SO24 − has already occurred and changes in the climate are more likely to affect the organic carbon concentration in the future. The comparison between the trend coefficients in SO24 − and TOC in this study did not indicate any relation even though the majority of the watercourses have a decrease in SO2− and an increase in TOC con4 centration. While the trend in TOC at the decadal scale has been remarkable, with major implications for assessment according to the European Union Water Framework Directive (Erlandsson et al., 2011), the monthly changes in mg l−1 resulting from discharge and seasonality were an order of magnitude larger than the trend. 4.4. Future applications This study presents a consistent empirical modeling approach to assess the environmental influences on stream TOC variability for multiple sites in a systematic manner. As shown by Temnerud et al. (2014), the method can also be applied to other water chemistry variables to facilitate identification of spatial similarities and differences in controlling processes. By including seasonal- and discharge-related effects, the method used here provides additional insights to intra-annual variability, compared to those obtained from long-term linear or monotonic trend analyses (e.g. Monteith et al., 2007). The method can also be used to investigate different future climate change scenarios for discharge and temperature assuming that the historical relations in the models remain constant under changed climate conditions. These statistical models can also serve as sentinels for changing catchment function that would be indicated by degradation in model performance when using calibrations on older data to model more recent data or even the future. 4.5. Conclusions The intra-annual variation in TOC concentration observed in watercourses throughout Sweden can primarily be explained by discharge and seasonality. This finding is of value for predicting TOC in the coming decades, but it also suggests where to focus efforts to further understand the processes controlling TOC. The regression models presented here facilitate systematic comparison of the influences of different terms on intra-annual variability over space and in time. Acknowledgments We thank Brian Huser for managing the soil type data, Claudia von Brömssen for statistical support, Asad Jamil for computing the theoretical water residence time from map data, and Greg Schwarz for the help in using Fluxmaster. Financial support for this research was provided by the Swedish University of Agricultural Sciences, the Swedish Environmental Protection Agency research project Climate Change and Environmental Objectives (CLEO Contract 09/115). Martyn Futter was supported by the MISTRA FutureForests program, the FORMAS ForWater strong research environment (grant number 230-2010-89) and the ECCO (funded by the Norwegian Research Council) and DomQua (funded by the Nordic Council) projects. Johan Temnerud was partly supported by the Swedish Meteorological and Hydrological Institute. Richard Alexander received support from the U.S. Geological Survey National Water Quality Assessment Program. The Swedish Environmental Protection Agency is responsible for funding the monitoring of the water quality data. We would also like to acknowledge the role of

J.K. Hytteborn et al. / Science of the Total Environment 520 (2015) 260–269

the Geochemical Laboratory at the SLU Department of Aquatic Sciences and Assessment for the consistent quality of the water chemistry data over half a century of monitoring. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.scitotenv.2015.03.041. References Ågren, A., Jansson, M., Ivarsson, H., Bishop, K., Seibert, J., 2008. Seasonal and runoff-related changes in total organic carbon concentrations in the River Öre, Northern Sweden. Aquat. Sci. 70, 21–29. http://dx.doi.org/10.1007/s00027-007-0943-9. Ågren, A., Haei, M., Köhler, S.J., Bishop, K., Laudon, H., 2010. Regulation of stream water dissolved organic carbon (DOC) concentrations during snowmelt; the role of discharge, winter climate and memory effects. Biogeosciences 7, 2901–2913. http:// dx.doi.org/10.5194/bg-7-2901-2010. Bossard, M., Feranec, J., Otahel, J., 2000. CORINE Land Cover Technical Guide — Addendum 2000. European Environment Agency, Copenhagen (105 pp. http://www.eea.europa. eu/publications/COR0-landcover). Boyer, E.W., Hornberger, G.M., Bencala, K.E., McKnight, D.M., 2000. Effects of asynchronous snowmelt on flushing of dissolved organic carbon: a mixing model approach. Hydrol. Process. 14, 3291–3308. Brown, M.B., Forsythe, A.B., 1974. Robust tests for the equality of variances. J. Am. Stat. Assoc. 69, 364–367. http://dx.doi.org/10.1080/01621459. 1974.10482955. Clark, J.M., Chapman, P.J., Heatwaite, A.L., Adamson, J.K., 2006. Suppression of dissolved organic carbon by sulfate induced acidification during simulated droughts. Environ. Sci. Technol. 40, 1776–1783. Clark, J.M., Bottrell, S.H., Evans, C.D., Monteith, D.T., Bartlett, R., Rose, R., Newton, R.J., Chapman, P.J., 2010. The importance of the relationship between scale and process in understanding long-term DOC dynamics. Sci. Total Environ. 408, 2768–2775. Dawson, J.J.C., Soulsby, C., Tetzlaff, D., Hrachowitz, M., Dunn, S.M., Malcolm, I.A., 2008. Influence of hydrology and seasonality on DOC exports from three contrasting upland catchments. Biogeochemistry 90, 93–113. de Wit, H.A., Mulder, J., Hindar, A., Hole, L., 2007. Long-term increase in dissolved organic carbon in streamwaters in Norway is response to reduced acid deposition. Environ. Sci. Technol. 41, 7709–7713. Eikebrokk, B., Vogt, R.D., Liltved, H., 2004. NOM increase in Northern European source waters: discussion of possible causes and impacts on coagulation/contact filtration processes. Water Sci. Technol. Water Supply 4 (4), 47–54 (http://folk.uio.no/rvogt/ CV/Co-author/NOM-increaseWtrSciTech-2004.pdf). Erlandsson, M., Buffam, I., Fölster, J., Laudon, H., Temnerud, J., Weyhenmeyer, G.A., Bishop, K., 2008. Thirty-five years of synchrony in the organic matter concentrations of Swedish rivers explained by variation in flow and sulphate. Glob. Chang. Biol. 14, 1191–1198. http://dx.doi.org/10.1111/j.1365-2486.2008.01551.x. Erlandsson, M., Cory, N., Fölster, J., Köhler, S., Laudon, H., Weyhenmeyer, G.A., Bishop, K., 2011. Increasing dissolved organic carbon redefines the extent of surface water acidification and helps resolve a classic controversy. Bioscience 61, 614–618. Fölster, J., Johnson, R.K., Futter, M., Wilander, A., 2014. The Swedish monitoring of surface waters: 50 years of adaptive monitoring. Ambio 43, 3–18. Fransson, S., 1965. The borderland. In: Sjörs, H. (Ed.), The plant cover of Sweden. Acta Phytogeographica Suecica 50. Almqvist & Wiksell, Uppsala, pp. 167–175 (http://uu. diva-portal.org/smash/get/diva2:565571/FULLTEXT01.pdf). Freeman, C., Evans, C.D., Monteith, D.T., Reynolds, B., Fenner, N., 2001. Export of organic carbon from peat soils. Nature 412, 785-785. Futter, M.N., de Wit, H.A., 2008. Testing seasonal and long-term controls of streamwater DOC using empirical and process-based models. Sci. Total Environ. 407, 698–707. http://dx.doi.org/10.1016/j.scitotenv.2008.10.002. Futter, M.N., Butterfield, D., Cosby, B.J., Dillon, P.J., Wade, A.J., Whitehead, P.G., 2007. Modeling the mechanisms that control in-stream dissolved organic carbon dynamics in upland and forested catchments. Water Resour. Res. 43, W02424. http://dx.doi. org/10.1029/2006WR004960. Futter, M.N., Forsius, M., Holmberg, M., Starr, M., 2009. A long-term simulation of the effects of acidic deposition and climate change on surface water dissolved organic carbon concentrations in a boreal catchment. Hydrol. Res. 40, 291–305. Gebrehiwot, S.G., Seibert, J., Gärdenäs, A.I., Mellander, P.-E., Bishop, K., 2013. Hydrological change detection using modeling: half a century of runoff from four rivers in the Blue Nile Basin. Water Resour. Res. 49, 3842–3851. http://dx.doi.org/10.1002/wrcr.20319. Geladi, P., Kowalski, B.R., 1986. Partial least-squares regression — a tutorial. Anal. Chim. Acta 185, 1–17. http://dx.doi.org/10.1016/0003-2670(86)80028-9. Grieve, I.C., 1991. A model of dissolved organic carbon concentration in soil and stream waters. Hydrol. Process. 5, 301–307. Ivarsson, H., Jansson, M., 1994. Regional variation of dissolved organic matter in running waters in central northern Sweden. Hydrobiologia 286, 37–51. Johansson, B., 2002. Estimation of Areal Precipitation for Hydrological Modelling in Sweden. Doctoral Thesis, University of Gothenburg, Gothenburg (36 pp.).

269

Jutras, M.-F., Nasr, M., Castonguay, M., Pit, C., Pomeroy, J.H., Smith, T.P., Zhang, C.-F., Ritchie, C.D., Meng, F.-R., Clair, T.A., Arp, P.A., 2011. Dissolved organic carbon concentrations and fluxes in forest catchments and streams: DOC-3 model. Ecol. Model. 222, 2291–2313. Karlsson, J., Byström, P., Ask, J., Ask, P., Persson, L., Jansson, M., 2009. Light limitation of nutrient-poor lake ecosystems. Nature 460, 506–510. http://dx.doi.org/10.1038/ nature08179. Köhler, S., 1999. Quantifying the Role of Natural Organic Acids on pH and Buffering in Swedish Surface Waters. Doctoral Thesis, Swedish University of Agricultural Sciences, Umeå (163 pp.). Köhler, S.J., Buffam, I., Laudon, H., Bishop, K.H., 2008. Climate's control of intra-annual and interannual variability of total organic carbon concentration and flux in two contrasting boreal landscape elements. J. Geophys. Res. 113, G03012. http://dx.doi.org/10. 1029/2007JG000629. Köhler, S.J., Buffam, I., Seibert, J., Laudon, H., 2009. Dynamics of stream water TOC concentrations in a boreal headwater catchment: controlling factors and implications for climate scenarios. J. Hydrol. 373, 44–56. Kullberg, A., Bishop, K.H., Hargeby, A., Jansson, M., Petersen, R.C., 1993. The ecological significant of dissolved organic carbon in acidified waters. Ambio 22, 331–337. Laudon, H., Berggren, M., Ågren, A., Buffam, I., Bishop, K., Grabs, T., Jansson, M., Köhler, S., 2011. Patterns and dynamics of dissolved organic carbon (DOC) in boreal streams: the role of processes, connectivity, and scaling. Ecosystems 14, 880–893. Ledesma, J.L.J., Köhler, S.J., Futter, M.N., 2012. Long-term dynamics of dissolved organic carbon: implications for drinking water supply. Sci. Total Environ. 432, 1–11. Lepistö, A., Futter, M.N., Kortelainen, P., 2014. Almost 50 years of monitoring shows that climate, not forestry, controls long-term organic carbon fluxes in a large boreal watershed. Glob. Chang. Biol. 20 (4), 1225–1237. Lindström, G., Bergström, S., 2004. Runoff trends in Sweden 1807–2002. Hydrol. Sci. J. 49, 69–83. http://dx.doi.org/10.1623/hysj.49.1.69.54000. Lindström, G., Pers, C., Rosberg, J., Strömqvist, J., Arheimer, B., 2010. Development and testing of the HYPE (Hydrological Predictions for the Environment) water quality model for different spatial scales. Hydrol. Res. 41, 295–319. Monteith, D.T., Stoddard, J.L., Evans, C.D., de Wit, H.A., Forsius, M., Högåsen, T., Wilander, A., Skjelkvåle, B.L., Jeffries, D.S., Vurenmaa, J., Keller, B., Kopácek, J., Vesly, J., 2007. Dissolved organic carbon trends resulting from changes in atmospheric deposition chemistry. Nature 450, 537–540. Monteith, D.T., Evans, C.D., Henrys, P.A., Simpson, G.L., Malcolm, I.A., 2014. Trends in the hydrochemistry of acid-sensitive surface waters in the UK 1988–2008. Ecol. Indic. 37, 287–303. Müller, R.A., Futter, M.N., Sobek, A., Nisell, J., Bishop, K., Weyhenmeyer, G.A., 2013. Water renewal along the aquatic continuum offsets cumulative retention by lakes: implications for the character of organic carbon in boreal lakes. Aquat. Sci. 75, 535–545. Reese, H., Nilsson, M., Granqvist Pahlén, T., Hagner, O., Joyce, S., Tingelöf, U., Egberth, M., Olsson, H., 2003. Countrywide estimates of forest variables using satellite data and field data from the National Forest Inventory. Ambio 32, 542–548. Runkel, R.L., Crawford, C.G., Cohn, T.A., 2004. Load estimator (LOADEST): A FORTRAN program for estimating constituent loads in streams and rivers. U.S. Geological Survey Techniques and Methods Book 4 (Chapter A5, 69 pp.). Sarkkola, S., Koivusalo, H., Laurén, A., Kortelainen, P., Mattsson, T., Palviainen, M., Piirainen, S., Starr, M., Finér, L., 2009. Trends in hydrometeorological conditions and stream water organic carbon in boreal forested catchments. Sci. Total Environ. 408, 92–101. Schwarz, G.E., Hoos, A.B., Alexander, R.B., Smith, R.A., 2006. The SPARROW surface water-quality model: theory, application, and user documentation. U.S. Geological Survey Techniques and Methods Report, Book 6. U.S. Geological Survey, Reston (Chapter B3, 248 pp. http://pubs.usgs.gov/tm/2006/tm6b3/PDF.htm). SGU. Soil type derived from database with scales 1:50000, 1:100000, 1:200000 and 1: 1.1000000, the best resolution thereof for each catchment. Data copyright by The Geological Survey of Sweden, Uppsala 2012. Sobek, S., Nisell, J., Fölster, J., 2011. Predicting the volume and depth of lakes from map-derived parameters. Inland Waters 1, 177–184. Strömqvist, J., Arheimer, B., Dahné, J., Donnelly, C., Lindström, G., 2012. Water and nutrient predictions in ungauged basins: set-up and evaluation of a model at the national scale. Hydrol. Sci. J. 57, 229–247. Temnerud, J., Hytteborn, J.K., Futter, M.N., Köhler, S.J., 2014. Evaluating common drivers for color, iron and organic carbon in Swedish watercourses. Ambio 43 (1), 30–44. Tukey, J.W., 1949. Comparing individual means in the analysis of variance. Biometrics 5 (2), 99–114. Winterdahl, M., Futter, M., Köhler, S., Laudon, H., Seibert, J., Bishop, K., 2011. Riparian soil temperature modification of the relationship between flow and dissolved organic carbon concentration in a boreal stream. Water Resour. Res. 47, W08532. Winterdahl, M., Erlandsson, M., Futter, M., Weyhenmeyer, G.A., Bishop, K., 2014. Intra-annual variability of organic carbon concentrations in running waters: drivers and relationships to long-term trends. Glob. Biogeochem. Cycles 28, 451–464. http://dx.doi.org/10.1002/¬2013GB004770. Yurova, A., Sirin, A., Buffam, I., Bishop, K., Laudon, H., 2008. Modeling the dissolved organic carbon output from a boreal mire using the convection–dispersion equation: importance of representing sorption. Water Resour. Res. 44, W07411. http://dx.doi.org/10. 1029/2007WR006523. Zeng, T., Arnold, W.A., 2013. Clustering chlorine reactivity of haloacetic acid precursors in inland lakes. Environ. Sci. Technol. 48 (1), 139–148. http://dx.doi.org/10.1021/ es403766n.

Patterns and predictability in the intra-annual organic carbon variability across the boreal and hemiboreal landscape.

Factors affecting total organic carbon (TOC) concentrations in 215 watercourses across Sweden were investigated using parameter parsimonious regressio...
3MB Sizes 1 Downloads 5 Views