Science of the Total Environment 518–519 (2015) 393–406

Contents lists available at ScienceDirect

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

Nutrient loadings from urban catchments under climate change scenarios: Case studies in Stockholm, Sweden Jiechen Wu, Maria E. Malmström ⁎ Industrial Ecology, Department of Sustainable Development, Environmental Science and Engineering (SEED), KTH Royal Institute of Technology, Teknikringen 34, SE-100 44 Stockholm, Sweden

H I G H L I G H T S • • • • •

Climate change effects on nutrient loadings in urban watersheds are investigated. A source model is integrated with a watershed model in a substance flow structure. Annual nitrogen loadings and the seasonal distribution may be modestly affected. Groundwater may potentially be the most sensitive pathway of nitrogen transport. Phosphorus loadings by water pathways may be less sensitive to climate change.

a r t i c l e

i n f o

Article history: Received 3 December 2014 Received in revised form 11 February 2015 Accepted 11 February 2015 Available online 13 March 2015 Editor: D. Barcelo Keywords: Urban catchments Diffuse sources Nutrient loadings Climate change Substance Flow Analysis

a b s t r a c t Anthropogenic nutrient emissions and associated eutrophication of urban lakes are a global problem. Future changes in temperature and precipitation may influence nutrient loadings in lake catchments. A coupling method, where the Generalized Watershed Loading Functions method was tested in combination with source quantification in a Substance Flow Analysis structure, was suggested to investigate diffuse nutrient sources and pathways and climate change effects on the loadings to streamflow in urban catchments. This method may, with an acceptable level of uncertainty, be applied to urban catchments for first-hand estimations of nutrient loadings in the projected future and to highlight the need for further study and monitoring. Five lake catchments in Stockholm, Sweden (Råcksta Träsk, Judarn, Trekanten, Långsjön and Laduviken) were employed as case studies and potential climate change effects were explored by comparing loading scenarios in two periods (2000–2009 and 2021–2030). For the selected cases, the dominant diffuse sources of nutrients to urban streamflow were found to be background atmospheric concentration and vehicular traffic. The major pathways of the nitrogen loading were suggested to be from both developed areas and natural areas in the control period, while phosphorus was indicated to be largely transported through surface runoff from natural areas. Furthermore, for nitrogen, a modest redistribution of loadings from surface runoff and stormwater between seasons and an increase in the annual loading were suggested for the projected future climate scenarios as compared to the control period. The model was, due to poor monitoring data availability, only able to set an upper limit to nutrient transport by groundwater both in the control period and the future scenarios. However, for nitrogen, groundwater appeared to be the pathway most sensitive to climate change, with a considerable increase and seasonal redistribution of loadings. For phosphorus, loadings by different pathways were apparently less sensitive to climate change. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Excessive nutrient loadings to water recipients can cause eutrophication problems, threatening aquatic environments and services provided to humans (e.g., Kaye et al., 2006). Furthermore, climate change has been projected to lead to changes in temperature and precipitation, ⁎ Corresponding author. E-mail address: [email protected] (M.E. Malmström).

http://dx.doi.org/10.1016/j.scitotenv.2015.02.041 0048-9697/© 2015 Elsevier B.V. All rights reserved.

factors affecting all parts of the hydrological cycle and causing e.g., flooding and changes in rainfall (De Risi et al., 2013; De Paola et al., 2013; Jalayer et al., 2014). These potential changes may influence nutrient loadings from catchments. Therefore, many models have been developed to investigate nutrient loading scenarios in response to climate change in order to combat future eutrophication. For example, the HBV model has been used for the Baltic Sea basin (Bergström and Graham, 1998), the SWAT model for the Vantananjoki watershed (Bouraoui et al., 2004), the Generalized Watershed Loading Functions

394

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

(GWLF) model for the Daugava river basin (Wallin, 2005) and the PolFlow model for the Norrström catchment and Lake Mälaren basin (Mourad and Van der Perk, 2004; Darracq et al., 2005). However, most previous studies have focused on large-scale catchments in a long-term view. Predictive models for urban catchments, where there are complex interactions between humans and environments, are still largely lacking. Such urban catchments are usually small in size and located in different areas of a city, but they play an important role in urban ecosystems. However, not all urban catchments are monitored sufficiently well to investigate future climate change effects on nutrient loadings. Therefore, studies of urban catchments are needed, particularly in areas characterised by a lack of basic data and previous studies, in order to highlight the need for further studies and to identify implications for monitoring in practice in the context of climate change. In order to handle future urban eutrophication from a nutrient flow management perspective, a clear understanding of nutrient sources, pathways and loadings in urban areas is also necessary. Coupling nutrient sources with the loadings to urban streamflow can be a useful first step in developing relevant scenarios addressing underlying changes in society in parallel with climate change. When investigating nutrient sources, diffuse sources to receiving waters are generally difficult to quantify and manage (e.g., Carey et al., 2013). In models dealing with urban drainage, e.g., the StormTac model in Sweden (Larm, 2000) and the SLAMM model in the US (Pitt and Voorhees, 2004), diffuse nutrient sources are often quantified together with other pollutants in stormwater through determining the pollutant concentration in different waters in combination with water flow data. Such models are capable of tracing source areas, but not the real primary sources. They are also rarely used to investigate climate change effects, as the dataset of projected temperature and precipitation in existing climate models cannot be directly plugged in. Although source-based approaches have been proposed for tracing primary sources of heavy metals (e.g., Sörme and Lagerkvist, 2002; Cui et al., 2010), one of the key aspects in these models is the fraction of pollutants delivered to different types of water paths. This is handled by an empirical factor that is unable to handle the consequences of future climate change. Furthermore, nutrient loadings from groundwater are not covered in these models. Therefore, accurate methods for prediction of urban diffuse nutrient sources, pathways and loadings through different water pathways in response to climate change are urgently needed. In a previous study, we qualitatively analysed climate change effects on nitrogen flows in Råcksta Träsk, an urban catchment in Stockholm, Sweden (Wu et al., 2013). Based on the results, we concluded that biological, hydrological, meteorological and biogeochemical effects and changes in human behaviour in response to climate change may lead to altered nutrient flows through urban catchments. The aim of the present study was to quantify nitrogen and phosphorus flows and their loadings under the future hydrological effects arising from climate change. Specific objectives were to: (1) identify major sources of nitrogen and phosphorus in urban streamflow and couple the sources with the loadings from selected catchments; (2) quantify the loadings of nitrogen and phosphorus contributed by different water pathways and identify the most sensitive pathway under future climate change; and (3) identify potential impacts from future climate scenarios on total loadings of nitrogen and phosphorus to urban streamflow.

of the results and consider the implications for future catchment management. In addition, they are isolated catchments with no upstream surface water system and are mainly supplied by streamflow from urban areas with various land uses, allowing potential urban nutrient sources and pathways to be investigated. In order to facilitate presentation and discussion of the results, the land uses were categorised into ‘developed areas’ and ‘natural areas’. The characteristics of the five lake catchments are shown in Table 1. In all five catchments, there is no heat and power production, refuse incineration or relevant industrial emissions that could cause airborne nitrogen (Malmqvist, 1983). The only local source of airborne nitrogen is traffic. Moreover, all five catchments largely comprise woodland and open land except Lake Långsjön, where 60% of the drainage area is composed of residential buildings. The drainage area of Lake Trekanten has the largest vehicular volume in its road area of the cases studied (Table 1). ‘Traffic area’ in the five cases includes road area, tram area and parking area. The catchment of Råcksta Träsk, in the western suburbs of Stockholm, includes a larger number of different land uses than the other catchments, with a horse riding area and construction area, categorised as the land uses ‘Livestock’ and ‘Contaminated soil’, respectively. These two land uses are not present in the other catchments. In addition, the drainage areas of lakes Råcksta Träsk, Trekanten and Laduviken have quite small areas of ‘Cultivated land’ that are near residential buildings, with negligible contributions of nutrients to urban runoff, and were thus excluded from the study. Therefore ‘Contaminated soil’ and ‘Cultivated land’ were lumped into the ‘Other permeable areas’ class, while ‘Livestock’ was taken as a separate land use given the particular source of horse droppings in the Råcksta Träsk area (Table 1). Moreover, there is a treatment pool and a lamella plant in the drainage area of Råcksta Träsk. They treat collected urban runoff and the treated water is discharged to the lake. Such facilities are not present in the other catchments studied. ‘Other permeable areas’ was placed under ‘Natural areas’ rather than ‘Developed areas’, given that they do not represent a land use dominated by impervious area as in the developed area category. We also quantified the sources using the same approach as for the natural areas, due to data availability. It should be noted that the original land use data were obtained from Stockholm Vatten (2000) with a high resolution of classification of land uses, and only the mentioned areas above are summed for purposes of presentation and discussion. The land use map for the five lake catchments is presented in Fig. 1, where the dedicated horse riding area is also noted. Despite the northerly location of Stockholm, it has a humid, continental climate and relatively mild weather compared with other locations at similar latitude. According to the Swedish Meteorological and Hydrological Institute (SMHI), the mean daily temperature in Stockholm in 2011 was 8.5 °C and mean annual precipitation was 478.8 mm (SMHI, 2014). SMHI maintains well-structured data series on daily temperature and precipitation for both historical periods and future climate projections for the Stockholm area (Kjellström et al., 2005). Potential climate changes can be expected in both the short and long terms (Stockholms Stad, 2007). Detailed climate data with a 20-year perspective were used in this study.

2. Case study

3.1.1. Nutrient flow analysis Substance Flow Analysis (SFA) was used to identify nitrogen and phosphorus flows within the five study catchments, which were generalized as shown in Fig. 2. The system boundary for each case was defined as the drainage area of the lake and the focus was on nutrient source identification, nutrient distribution for each land use type and loadings to streamflow. Streamflow in this work refer to water flows caused by rainfall and snowmelt that can potentially transport nutrients into water

Five lake catchments in Stockholm, Sweden (Råcksta Träsk, Judarn, Trekanten, Långsjön and Laduviken), which are regularly monitored for water quality, were used as case studies in this work. These five catchments were selected because they represent different proportions of various land uses and different levels of nutrient concentrations in lake water (Stockholm Vatten, 2006), in order to allow generalization

3. Methods 3.1. Overview of the approach

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

395

Table 1 Characteristics of the five urban catchments in Stockholm, Sweden, used as case studies in this work. Abbreviation

Land usesa

Råcksta Träsk

Judarn

Trekanten

Långsjön

Laduviken

19% 2% 13% 18% 0%

11% 0% 1% 61% 0%

9% 0% 13% 2% 0%

28% 20% 1% 60.2 61d 0.5 3.6

7% 19% 0% 241.7 1.5d 23.8 1.9

42% 32% 2% 99.8 17e 0.7 2.4

Fraction of total catchment Developed areas TR IN BU RE LS Natural areas OL WL OP Traffic

a b c d e

Traffic area Industrial area Business area Residential area Livestock

9% 5% 12% 5% 1%

Open land Woodland Other permeable areas Totala (ha) Vehicles (1,000,000 vehicles/yr) Road lengtha (km) Parking areaa (ha)

31% 32% 7% 340.9 1.2b 1.7 5.8

9% 0% 1% 2% 0%

23% 65% 0% 74.2 9.5c 1.6 0.3

Stockholm Vatten (2000). Stockholms Stad (2006). Miljöförvaltningen (2002). Larm and Holmgren (1999). Miljöförvaltningen (2003).

recipients. They were categorised as stormwater, surface runoff and groundwater, which corresponded to the water flowing from urban developed areas, natural areas and the saturated zone in the soil, respectively. Based on previous field and literature studies of nutrient sources in streamflow in Swedish urban catchments (e.g., Malmqvist, 1983; Wu et al., 2013), the sources identified in this study were: background atmospheric concentration (Source I), vehicular traffic (Source II), parking (Source III) and horse droppings (Source IV) for developed areas; and vegetation-related sources (e.g., atmospheric concentration, organic litter, nitrogen fixation and soil leaching) (Source V) for natural areas (Fig. 2). The separate categorisations for developed areas and natural areas were made given data availability and the main objective of this work. For background atmospheric concentrations in particular, the estimates used were only for developed areas. It is possible to quantify background atmospheric concentration for natural areas, but quantification of other sources such

as organic litter, soil leaching, etc. would require detailed data. Due to lack of data and our particular interest in human-dominated areas, a simple concentration-based approach was applied to estimate lumped vegetation-related sources in natural areas, where background atmospheric concentrations are already taken into account. As shown in Fig. 2, nutrient distribution was defined as the amount of nutrients apportioned from primary sources to each land use, where the inflows were quantified before being entrained by water pathways. The loadings from each land use were taken as the amounts of nutrients entering streamflow during rainfall and/or snowmelt events. The nutrients in stormwater and surface runoff were assumed to be delivered by stormwater pipes or open ditches to water recipients or treatment sectors, according to the different drainage systems for each case. Nutrients in groundwater were assumed to be delivered to water recipients.

Fig. 1. Position of catchments in Stockholm, Sweden, and land uses in the five urban catchments, used as case studies in this work. Land use abbreviations: TR: Traffic area; IN: Industrial area; BU: Business area; RE: Residential area; LS: Livestock; OL: Open land; WL: Woodland; OP: Other permeable areas.

396

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

Fig. 2. Conceptual nutrient flow analysis of the drainage area of the five lake catchments. The dotted lines indicate the flows that are only applicable to nitrogen; the thick grey arrows represent water flows and sediment yield, and the thin black arrows represent associated nutrient flows. The ‘S’ and ‘C’ labels distinguish different quantification methods in the model, where S represents source-based nutrient inputs and C represents concentration-based nutrient inputs. Land use abbreviations: TR: Traffic area; IN: Industrial area; BU: Business area; RE: Residential area; LS: Livestock; OL: Open land; WL: Woodland; and OP: Other permeable areas. (Inspired by Cui et al., 2010; Wu et al., 2013).

3.1.2. Modelling strategy for streamflow loads Using the SFA structure identified, we first quantified the nutrient sources and then coupled the quantifications with a GWLF model, which was chosen due to its wide use for climate change effect studies and its engineering compromises between empiricism and complexity of chemical simulations (Schneiderman et al., 2010). The coupled model was calibrated and tested through comparisons with independent nutrient loading estimations from the literature for the current situation. Finally, in order to investigate climate change effects, only changes in temperature and precipitation were considered and the projected data were taken directly from state-of-the-art climate models in the literature. Source strengths and land use, as well as the inflows to each land use type, were assumed to remain constant when assessing the effects of climate change on nutrient loadings. The GWLF simulations were implemented through Visual Basic for Applications (VBA) running in Microsoft Excel (GWLFXL 2.1.1; Hong and Swaney (2007)). 3.2. Quantification of source strengths As already mentioned, a source-based approach was employed to quantify source strengths for developed areas, while a concentrationbased approach was used as a supplementary method for natural areas due to poor data availability. The source-based approach was based on the assumption that leaching is the dominant emission mechanism (Elshkaki et al., 2005). The emissions (E, kg/year) from sources were quantified as: Em;n ðt Þ ¼ μ m Sm;n ðt Þ

ð1Þ

where Sm,n(t) is the size of the stock of source, m, in land use, n, at time, t (=1 year), and μm is the leaching factor. The parameters set for the source model are shown in Table 2. For vehicular emissions from the road area, vehicles were assumed to contribute airborne nitrogen and particulate phosphorus. These emissions of airborne nitrogen from the road area (not relevant industrial emissions in this study) formed part of total atmospheric nitrogen concentration and were calculated using Eq. (2) in Table 2. The corresponding emissions of particulate phosphorus were calculated using Eq. (6) in Table 2. The particulate phosphorus was assumed to accumulate on road surfaces and wash off directly to stormwater during storm events. For horse droppings in the livestock area, nutrient emissions were calculated using Eq. (3) in Table 2. For background atmospheric concentration (calculated using Eq. (4) in Table 2), area a (ha) was considered the stock and accumulation rate r (kg/ha & year) the leaching factor.

For the parking area and all the vegetation-related primary sources, source strength was estimated using a concentration-based approach in which nutrient loadings from natural areas were calculated by multiplying specific area and specific concentration of surface runoff by annual precipitation, assuming that primary water flow due to precipitation occurred at a time before evapotranspiration and soil absorption etc. This assumption was used to identify the magnitude of their contribution to the total nutrient loadings and then to evaluate the necessity of investigating primary sources in future studies, using Eq. (5) in Table 2. All the parameters and input data used for source quantification can be found in Tables 1 and 2. 3.3. Linking source quantifications to the GWLF model 3.3.1. GWLF model The GWLF model dynamically simulates the hydrological cycle in a watershed, taking evaporation, rainfall and snowmelt into account and predicting runoff from different land uses, groundwater, soil erosion and sediment yield (see thick arrows in Fig. 2). Nutrient loadings associated with these flow paths and yield sediments (see thin arrows in Fig. 2) are estimated by empirically derived specific sources or specific concentrations for different land uses (Haith et al., 1992). Biogeochemical processes of nutrients are not simulated in GWLF, but since the present study aimed to provide a first approximation of the potential hydrological effects of climate change on nutrient loadings, the role of hydrology was of most concern, in transporting nutrients to water recipients (Moore et al., 2008). The GWLF model is driven by daily mean temperature and daily precipitation. Besides the input from weather data, nutrient input and catchment characteristic-related transport data are also required. Water balances and associated nutrient loadings are simulated with a daily time step. The simulated daily results can be summed by month or year for nitrogen and phosphorus (dissolved and total) loadings to runoff from each land use area, as well as loadings through groundwater. Note that the GWLF model can also predict nutrient loadings from point sources and septic tank systems, which were not included in this study due to the nature of the cases selected and the aims of the work. In addition, model output on water flows and sediments is not presented, as the focus was on nutrient flows. 3.3.2. Source quantification as input to the GWLF model The nutrient data inputs are the link between source quantification and the GWLF model. For developed areas, GWLF estimates nutrients based on the build-up and wash-off mechanism proposed by Amy (1974), where nutrients accumulate on the surface over time at a

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

397

Table 2 Parameters used in the model and value setting for source quantification. Em,n(t), (kg/day) Nitrogen EII = NTLwv … (2)

Vehicular traffic EIII = nmd … (3)

Horse droppings EIV = ar …(4)

Background atm. EV = (aCsP) …(5)

Parking Vegetation

Phosphorus EVI = NTLwv … (6)

Vehicular traffic EIII = nmd …(3)

Horse droppings EIV = ar …(4)

Background atm. EV = (aCsP) …(5)

Parking Vegetation

a b c d e f g h i

Leaching factor, μm

Stock, Sm,n(t)

N emission rate N mg/vehicle km 1.1a Number n Horses 50d Area a ha

Length L km

Volume T Vehicles/yr

b

b

Production rate m kg/horse-year 48e

b

Area a ha

Concentration Cs mg/l 1.1h 1h 0.6h 1.8h

b

Open land Woodland Other permeable areas

b

Emission rate V mg/vehicle km 0.6h Number n Horses 50d Area a ha

Length L km

Volume T Vehicles/yr

b

b

b b

Production rate m kg/horse-year 16e

b

Area a ha

Concentration Cs mg/l 0.1h 0.03h 0.02h 0.13h

b

Open land Woodland Other permeable areas

b b b

Ratio to total atm. fallout w % 19c Loss rate d % 0.3f Accumulation rate r kg/ha-year 16g Annual precipitation ppn. cm 54i 54i 54i 54i

Paved area v % b

After street sweeping c % 90h Loss rate d % 0.7f Accumulation rate r kg/ha-year 0.64h Annual precipitation ppn. cm 54i 54i 54i 54i

Johansson and Eneroth (2007). Values for the five catchments are shown in Table 1. Bettez et al. (2013). Data obtained from the riding school. Lundgren and Pettersson (2004) and Barles (2007). Karlsson and Torstensson (2003). Malmqvist (1983). Larm (2000). SMHI (2014). ppn is mean annual precipitation for the hydrological year 2000-2011.

of sources is uniform over the course of a day (Haith et al., 1992). Nutrient input to each land use (Fig. 2) is traced back to different primary sources that are quantified on an annual basis due to data availability, with units of kg/ha & year. When linking to GWLF, the annual input for specific land use is suggested to be averaged to daily input, converting the units from kg/ha & year to kg/ha & day with an even distribution over the year. For natural areas in GWLF, nutrient loadings come from sediments by soil erosion and dissolved nutrients in surface runoff, where the model needs soil data and the specific concentration of nutrient for

constant accumulation rate and a constant depletion rate during dry periods, and are washed off by runoff events. Some noteworthy assumptions made for this mechanism in the present study were that accumulated nutrient load on the source area reached 90% of the maximum in 20 days and that 1.27 cm of runoff could wash off 90% of accumulated nutrients. These values were based on previous field studies (Haith et al., 1992). As GWLF simulates nutrient loadings with a daily time step, it requires a daily nutrient input rate to each type of land use (Hong and Swaney, 2007). It is assumed in GWLF that the distribution

Table 3 Climate models and scenarios used for future projections.

Climate scenario GCM RCM Emission scenarios from IPCC

C A1

E A1

E A2

H A1

CCSM3_A1B CCSM3 RCA3 A1 B⁎

ECHAM5_A1B_3 ECHAM5 RCA3 A1 B

ECHAM5_A2_1 ECHAM5 RCA3 A2⁎

HADCM3_A1B HADCM3 RCA3 A1 B

⁎ Note: A1B and A2 are explained in detail in Nakicenovic et al. (2000).

398

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

the land uses. Dissolved nutrient loadings are estimated by multiplying a land use-specific nutrient concentration by predicted runoff, while solid phase nutrient loadings are estimated by multiplying average sediment nutrient concentrations by predicted sediment yields. Here, land use-specific nutrient concentrations (mg/l) were included in the source quantifications (Table 2). Runoff is calculated by GWLF using the SCS curve number method (Ogrosky and Mockus, 1964). The nutrient transport by groundwater in GWLF is determined by the amount of groundwater discharge and the nutrient concentration in groundwater, and it is assumed that nutrient concentration remains constant (Haith et al., 1992). Thus both the quantities of nutrient leaching to groundwater and the quantities of nitrogen returning to the atmosphere through the denitrification process are not particularly simulated by GWLF. For catchment characteristic-related transport data in the GWLF model, land use data were based here on the land use classes of Stockholm (see Fig. 1), but with some land uses lumped depending on their proportion in the selected catchments and an appropriate interpretation for urban catchments (see Table 1). Curve numbers and parameter values for the Universal Soil Loss Equation (USLE), a sub-function used in GWLF to calculate soil erosion, were taken from the literature (Haith et al., 1992; Dai et al., 2000; Wallin, 2005; Hong and Swaney, 2007). Application of weather data is described in Section 3.4. Detailed nutrient data and transport data are provided in Appendix A.

Fig. 3. (a) Mean monthly air temperature and (b) mean monthly precipitation under the four climate scenarios. Black curves refer to the control period 2000–2009 and other curves to the future period 2021–2030. In the control period, mean daily temperature was 8 °C and annual mean precipitation was 54.46 cm, while changes in future periods were: Temperature (°C), C A1 (−0.87) H A1 (−0.89) E A2 (−1.13) E A1 (−1.32); precipitation (cm), C A1 (+25.10) E A2 (+21.25) E A1 (+18.68) H A1 (+16.91).

3.4. Climate scenarios To evaluate climate change effects on nutrient loadings, two 10-year periods were used: a control/reference period (2000–2009) and a future period (2021–2030). Climate data for the control period comprised observed daily temperature and precipitation (SMHI, 2014). The future daily temperature and precipitation data, with 50 km × 50 km field resolution, were obtained from SMHI (SUDPLAN, 2014). There were four available series of climate scenario data for the Stockholm region and all these were used to handle uncertainties and bias in the climate projections. These four future climate scenarios are based on the regional climate model Rossby Centre Atmosphere (RCA3), using boundary conditions from three global models (Community Climate System Model Version 3 (CCSM3); the fifth generation ECHAM general circulation model (ECHAM5); and a third version of the Hadley Centre coupled

Fig. 4. Modelled annual loading of (a) nitrogen and (b) phosphorus through surface runoff and stormwater from all land use areas in the base case of this study versus loadings estimated by Stockholm Vatten (2000). For both diagrams, grey and dark markers indicate results before and after calibration, respectively. The diagonal line indicates full agreement between the modelled and reported estimates. Note: Nutrient loadings by groundwater are not taken into account in Stockholm Vatten (2000) and are thus excluded here (modelled and reported data).

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

model (HADCM5)) and two IPCC emission scenarios (A1B and A2) (Table 3). The A1B emission scenario depicts an integrated future world and scenario A2 a heterogeneous future world in terms of population, technology and socio-economics (Nakicenovic et al., 2000). In the Stockholm region for the future period 2021–2030, mean monthly temperature decreased in these four scenarios and annual precipitation increased compared with the control period (Fig. 3). In order to discuss climate change effects on monthly and yearly nutrient loading, model results from the two periods were compared. For each period, the daily nutrient loading over 10 years was delivered as model output. For each month and year, the model output was aggregated into sets of 10 samples corresponding to each of the years in the period. For reporting and comparing the results of the control and future scenario periods, the statistics (maximum, minimum, median, interquartile range) for the 10 samples in the two sets were used. 3.5. Model calibration and testing No data were available for testing the SFA and GWLF model separately. Independent estimates of nutrient loading by stormwater and surface runoff from four of the lake catchments in our study (Råcksta Träsk, Judarn, Trekanten and Laduviken) were found in the local official literature (Stockholm Vatten, 2000), but only for one year, 2000. Stockholm Vatten (2000) estimated the loadings based on water volumes draining from each land use area multiplied by the standard concentration of nutrients for that specific land use. For the fifth case, no independent information was available. Note that groundwater is not taken into account in Stockholm Vatten (2000). These data were used for model calibration and testing for nutrient loadings by stormwater and surface runoff. A discussion on nutrient loadings by groundwater is found in Section 3.6. It should be noted that the GWLF model usually uses streamflow data to calibrate the model (e.g., Moore et al., 2008) if sufficient monitoring data exist for the studied catchment. This was not the case for the five lake catchments in this study. Thus we simulated the streamflow for the selected catchments (data not presented) and used resulting nutrient loadings for calibration and model testing in conjunction with the literature data on nutrient loadings. In order to allow the reported results to be compared with our model results, the modelled loadings through groundwater were excluded in the comparison between output from our model and data from Stockholm Vatten (2000). The grey markers in Fig. 4 show the modelled nutrient loading from the combined SFAGWLF model (vertical axis) as a function of independent reported results from Stockholm Vatten (2000) for the four cases (Lakes Judarn, Trekanten, Laduviken and Råcksta Träsk) in 2000 before calibration. The line in Fig. 4 shows full agreement between the two estimates. To assess the correlation (goodness-of-fit) between the model output and the independent literature values, we calculated NashSutcliffe Efficiency (NSE) (Nash and Sutcliffe, 1970). While NSE is usually used to assess observed and predicted values, here we used the reported results from Stockholm Vattern (2000) as nominal observed values, as no field data were available. NSE is calculated as NES ¼ 1:0− T

2

T

∑t¼1 ðyt −f t Þ =∑t¼1 ðyt −yÞ2 , where yt is the observed data value at time t, ft is the model simulated value at time unit t, and t = 1, 2,…, T. y is the mean of observed data value. NSE indicates how well the plot of the observed value versus the simulated value fits the 1:1 line, and ranges from −∞ to 1 (Nash and Sutcliffe, 1970). The larger the NSE value, the better the model performance. For phosphorus (Fig. 4b), the agreement between modelled and reported literature values was very good for all four cases (see the grey markers; NSE = 0.982), even before calibration. For nitrogen (Fig. 4a), the modelled values were somewhat lower than the values reported by Stockholm Vatten (2000), but showed similar behaviour between the cases, with Råcksta Träsk having much greater and Judarn somewhat smaller nutrient loadings than the other two cases (see the grey

399

markers; NSE = 0.769). Based on that our coupled SFA-GWLF model can (even prior to calibration) produce estimates that are comparable to those from other existing models, we suggest that the model can be used for first-hand nutrient accounting and projections in future scenarios. While there is no evidence on whether the estimates from the coupled SFA-GWLF model or the simpler model used by Stockholm Vatten (2000) are more accurate, for the sake of utility in prediction of future loadings (as affected by climate change) we calibrated our model against the nutrient loadings from the results reported (as nominal observed values) by Stockholm Vatten (2000). Nitrogen loadings by stormwater and surface runoff from the lake catchment of Råcksta Träsk were selected for the calibration process, given that a larger difference existed in nitrogen loadings between the simulated results (before calibration) and the reported results in that case, and also that this lake catchment included the highest number of land use types among the five lake catchments studied. Since the nitrogen loadings from each land use could be obtained from both Stockholm Vatten (2000) and our model results, this calibration was achieved by tracking nitrogen loadings from each land use type and adjusting the curve number in the GWLF subroutine for woodland and open land until a match was achieved (see the black marker for Råcksta Träsk in Fig. 4a and the calibrated value shown in Appendix A). The calibrated model was tested for phosphorus loading for Råcksta Träsk and nitrogen and phosphorus loadings for the other three lakes, by comparing the simulated results with the reported results. Note that these results do not include loadings by groundwater. The black markers in Fig. 4 show modelled loadings after the model calibration (vertical axis). As expected, calibration led to a better match between previously reported values and results obtained from the combined SFA-GWLF model for both nitrogen (NSE = 0.998) and phosphorus (NSE = 0.992). We deemed this level of agreement to be acceptable and used the calibrated model for further investigations. 3.6. Sensitivity and uncertainty analyses The sensitivity of model results to different parameters was investigated for both the source model and the combined SFA-GWLF model. In order to understand the quality of the modelled results, the uncertainty in model input for the parameters and how this was reflected in the model output were also assessed for both the source model and the combined SFA-GWLF model. In particular, the performance of uncertainty analysis for the combined model was measured using the coefficient of variation (CV), defined as SD/MV (standard deviation/mean value). Details of these analyses can be found in Appendix B. Model calibration and testing in Section 3.5 excluded loadings by groundwater due to poor data availability. Nutrient loadings by groundwater are estimated in GWLF through multiplying specific nutrient concentration in groundwater (defined and kept constant by GWLF) and the simulated groundwater discharge. In the GWLF model, the daily water balance is calculated for unsaturated and shallow saturated zones. Percolation from unsaturated to shallow saturated zones occurs when the unsaturated zone exceeds the field capacity (U*). In simulating the water budget for the shallow saturated zone, groundwater discharge and deep seepage are determined in the model by groundwater recession (r) and seepage constant (s), respectively (Haith et al., 1992). In favourable cases, streamflow data can be used to calibrate the model by adjusting U*, r, and/or s. However, for our cases, monitoring data for such calibration were not available. In our model implementation, we used the GWLF software's default value of U* (U* = 10 cm) and a typical value of r (r = 0.01/day) from Dai et al. (2000). For the seepage constant (s), in the model description of GWFL, the default value is 0, meaning that there is no water seepage from the shallow to the deep saturated zone, potentially causing overestimation of the groundwater discharge. This default use of s = 0 is called the “conservative approach” (Haith et al., 1992). The seepage constant

400

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

Fig. 6. Nutrient loadings of (a) nitrogen and (b) phosphorus by water pathways (excluding groundwater) from different source areas for Råcksta Träsk for the control (2000–2009; open boxes) and the C A1 climate scenario (2021–2030; filled boxes), with source strengths kept constant in the two periods. Box plots indicate median and interquartile ranges, and whiskers show minimum and maximum values in every 10 simulations. Land use abbreviations: TR: Traffic area; IN: Industrial area; BU: Business area; RE: Residential area; LS: Livestock; OL: Open land; WL: Woodland; OP: Other permeable areas.

Fig. 5. Yearly contribution of the different diffuse sources of (a) nitrogen and (b) phosphorus, normalised to catchment size, in the drainage area of five lake catchments in Stockholm, Sweden, according to the source quantifications (baseline; SFA model).

can only be calibrated in the model from available streamflow data, since the seepage constant is difficult/impossible to monitor in the field (Haith et al., 1992). In view of the lack of monitoring data to estimate or calibrate the seepage constant for use in the GWFL model, we used the conservative approach and set s = 0/day (Haith et al., 1992). This implies that there is no water flow to the deep unsaturated zone in the model, and thus the estimated groundwater discharge and nutrient transported by groundwater represent upper estimates (flows at their maximum). In order to examine how the estimate of groundwater discharge may be affected by using this conservative approach, the sensitivity of annual groundwater discharge to the seepage constant was investigated (see Appendix B). In addition, the sensitivity of groundwater discharge to the other parameters and the uncertainty of the estimated groundwater discharge were also investigated (see Appendix B). 4. Results and discussion 4.1. Source quantification The contributions of different diffuse sources of nitrogen and phosphorus in the drainage area of the five lake catchments in Stockholm, Sweden, according to the source quantifications (SFA model) after normalisation to total catchment area are shown in Fig. 5. For all cases, the main nutrient sources were background atmospheric concentration (black bars in Fig. 5), vehicular traffic (dark grey bars), and vegetationrelated sources (dotted bars), but these varied in importance between the cases. It should be noted that background atmospheric deposition

was included in the vegetation category for natural areas here, but handled separately for developed areas. Thus, the apparent catchment size-normalised atmospheric deposition was larger for cases with a larger proportion of developed land (Långsjön, Trekanten and Råcksta Träsk; Table 1). Nitrogen and phosphorus from vehicular traffic reflect traffic volume. For instance, the dominant source in the Trekanten catchment was vehicular traffic due to the highly trafficked motorway within the catchment. For all cases, parking and horse droppings were small sources compared with other sources (Fig. 5). The dominant role of atmospheric concentration and vehicle traffic within urban catchments is in agreement with Sommer et al. (2008), who calculated sources and fluxes of nutrients for an urban catchment in Germany and concluded that nitrogen originated from atmospheric deposition and most phosphorus from vehicular traffic and atmospheric deposition. As indicated in Fig. 5, although four out of five lake catchments studied here had a dominant proportion of natural areas (about 70%; see Table 1), sources in developed areas (background atmosphere, vehicular traffic, parking and horse droppings) made more noticeable contributions and in some cases were even significantly higher than the vegetation-related sources, e.g., for Lake Trekanten. As explained in Appendix B, the overall results of the source quantification were most sensitive to parameters quantifying the dominant sources. The dominant source was case-dependent, but in general, the three major sources were background atmospheric deposition, vehicular traffic emissions and vegetation-related sources. Taking phosphorus emissions in Råcksta Träsk as an example, the largest source was background atmospheric deposition and thus the most sensitive parameters were found to be the atmospheric deposition rate (r) and the investigated area (A), which quantify this source. In addition, based on uncertainty propagation, the uncertainty of the individual nitrogen and phosphorus source estimates was 13–17% for Råcksta Träsk, Judarn and Laduviken and 18–39% for Trekanten and Långsjön, thus falling within the definition of high quality data for SFA based studies, given as uncertainty less than 100% by Bi et al. (2013).

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

401

Fig. 7. Yearly loadings of (a–b) nitrogen and (c–d) phosphorus to Råcksta Träsk in the control period (2000–2009) and the future climate scenario period (2020–2030; based on projected weather data shown in Fig. 3, climate models are explained in Table 3), with source strengths kept constant in and between the two periods. Box plots indicate median and interquartile ranges, and whiskers show minimum and maximum values in every 10 simulations (10 years in each period). On the left side of (a) and (c) show total nutrient loading from stormwater and surface runoff; and the right side of (a) and (c) show dissolved nutrient loading from stormwater and surface runoff. (b) and (d) show total nutrient loadings from groundwater (only dissolved nutrients as defined in the GWLF-model).

4.2. Nutrient loadings from different land uses The modelled, median yearly loadings of nutrients to Råcksta Träsk through water paths from different land uses (excluding groundwater) for the control period (open markers; 2000–2009) and for the climate scenario C A1 (filled markers; 2021–2030) are shown in Fig. 6. The modelled loadings for Råcksta Träsk in the other climate scenarios showed similar patterns (not shown), and the modelled results for the other catchments were also similar (not shown). The simulated yearly nutrient loading was obtained by integrating daily delivery of nutrients from the combined SFA-GWLF model, with the climate scenario reflecting predictions of future temperature and precipitation (Fig. 3), but not including other changes in the catchment over time (such as changes in land use and nutrient source strengths). For the present situation (2000–2009), open land (“OL” in Fig. 6) and business areas (“BU”) gave the largest delivery of nitrogen, while other permeable areas (“OP”) and open land gave the largest delivery of phosphorus. The results suggest that in the control period, the major pathways of the nitrogen loading could be from both developed areas and natural areas: about 55% of the nitrogen loadings from different land uses

were transported by stormwater (from all the developed areas) and about 45% were transported by surface runoff (from all the natural areas). For phosphorus, the loadings were instead predominantly from surface runoff (78%). For both nutrients, the response to climate change (assuming Scenario C A1) was minor for all natural land use types (surface water), but there was a predicted slight increase in nitrogen loading from all developed land use by stormwater. The model results suggest that the amount of nutrient loading by stormwater was greater in C A1 than in the control, due to increased stormwater from developed areas. The change in nutrient loadings by surface runoff from natural areas was not as obvious as that by stormwater, since runoff and sediment erosion from natural areas were not found to undergo obvious changes. The results obtained for the other lake catchments were similar to those for Råcksta Träsk (see Fig. 6). An exception was Långsjön, for which the modelling results showed increased phosphorus loadings by stormwater from developed land uses. This can be explained by the fact that this catchment has the largest developed area (see Table 1) and the hydrology is dominated by impervious surface area, making it sensitive to climate change effects (data not shown).

402

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

4.3. Annual nutrient loadings Fig. 7 shows simulated yearly median integrated nutrient loadings by water pathways for Råcksta Träsk for the control period and future climate period, including integrated effects of changes in temperature and precipitation. The results for the other lakes were similar (not shown). The integrated loadings in Fig. 7a and c were obtained by aggregating results from all the land use types (stormwater and surface runoff also shown in Fig. 6), where markers to the left of the dotted line represent total loading and those to the right represent the loading of dissolved nutrients, the difference being made up by transported particulate nutrients. The loadings from groundwater are separately presented in Fig. 7b and d, where the loadings constitute only dissolved nutrients, as defined by GWLF. For the control period, the transport of dissolved and particulate nitrogen (indicated by difference between “Tot. N” and “Dis. N” in Fig. 7a) was roughly equally important for the loading of nitrogen by stormwater and surface runoff (open markers in Fig. 7a). For all future climate scenarios, the loading of total nitrogen from stormwater and surface runoff was somewhat higher than in the control period, while for the loading of dissolved nitrogen, there was no noticeable difference between the two periods. This suggested that the increase of total nitrogen loadings was mainly from particulate nitrogen, which corresponds to the results in Fig. 6a: noticeable increases occurred for nitrogen loadings from the developed land uses (stormwater), and as defined by GWLF, the nitrogen loadings from developed land uses are in the form of particulate nitrogen. No noticeable difference in loading of phosphorus between the control period and the future climate scenarios was indicated by the model results (Fig. 6c). Although phosphorus transport by stormwater increased with increasing stormwater amount, this did not greatly affect the overall total loadings from stormwater and surface runoff. This is because the loadings from open land and other permeable areas contributed the largest proportions, yet did not show noticeable changes in future change scenarios (see Fig. 6b). Fig. 7b shows the upper limit of nitrogen transported by groundwater. In contrast to nitrogen transported by stormwater and surface runoff (Fig. 7a), model results indicated a dramatically increased groundwater transport of nitrogen in the future scenarios compared to the control period. In addition, the modelled nitrogen loading by groundwater varied between the different studied climate scenarios. In Fig. 7b, the scenarios are organised from left to right in order of decreasing response to climate change (C A1, E A1, E A2 and H A1; compare Fig. 3). Since the nutrient concentration in groundwater was kept constant (as defined by GWLF) when simulating the control and climate scenarios, the increased nutrient loadings transported by groundwater are connected to increased groundwater flow in the future climate scenarios compared with the control period. Scenario C A1, which included the largest increase in groundwater discharge, thus, had the greatest effect on nitrogen loading by groundwater of all of the scenarios. Little phosphorus was transported by groundwater in the control period (see Fig. 7d). Due to the lower phosphorus than nitrogen concentration in groundwater, the increase in groundwater transport in response to climate change was not noticeable for phosphorus with the scale in Fig. 7d, although a substantial increase in groundwater discharge was predicted for the different future climate scenarios. For the considerable increase in groundwater discharge, it should be noted that the estimated groundwater discharge by GWLF in this work represents an upper limit, due to the conservative approach used (see Appendix B). Thus, the nutrient transport by groundwater was also the maximum that may potentially be expected with different scenarios (discussed in Section 3.6). For the uncertainty in groundwater estimates, the CV was suggested to average at 0.50, based on 100 realisations for the period 2000–2009 (see Appendix B). Taking Råcksta Träsk as an example, the annual groundwater discharge was most sensitive to land use-specific curve numbers, particularly for the land uses: open land, woodland and business area. Furthermore, seepage constant was suggested as an important factor in

simulating groundwater discharge under future climate scenarios, since it was found that the groundwater discharge was highly sensitive to the value of seepage constant, particularly at low seepage (see Appendix B). Despite the uncertainty in the actual groundwater discharge estimates and associated nutrient loadings through groundwater, a predicted increase of nitrogen loading by groundwater from the control period to the future climate scenario is robust with respect to uncertainties in groundwater discharge modelling in this study (see Appendix B). Thus groundwater transport is a sensitive pathway that may potentially cause considerable increase in nitrogen loading. It may also need particular attention in the context of climate change in terms of data need in future research and monitoring needs for future flow management. The combined nitrogen loadings from the three water pathways (stormwater, surface runoff and groundwater in Fig. 7a and b) implied a potentially substantial increase in total nitrogen loadings due to climate effects on temperature and precipitation in our future scenarios, which agrees with findings reported by Darracq et al. (2005). They applied the PolFlow model to formulate nutrient loading scenarios for the Norrström catchment west of Stockholm (2.2 million ha; regional scale as opposed to our local scale) and predicted that the nitrogen loading at the outlet of the basin by 2030 would be greater than at present, due to climate change effects on hydrometeorological characteristics. Taking Råcksta Träsk as an example, the median change in phosphorus loading due to climate change in the present study ranged from −4.3% to + 17% between the control period and the future scenario period, while Darracq et al. (2005) found a 0.04% increase in phosphorus loading from the Norrström catchment by 2030. Although the differences are not identical between our study and theirs, the effect of climate change on phosphorus loading was negligible to modest according to both models and over all cases. The differences between model predictions for the cases might stem from differences in geographical scale, time period, land use or model formulation of sources and transport

Fig. 8. Mean monthly loadings of (a) nitrogen and (b) phosphorus to Råcksta Träsk for the control (2000–2009; open boxes) and C A1 climate scenario period (2020–2030; filled boxes), with source strengths kept constant in the two periods. Box plots indicate median and interquartile ranges, and whiskers show minimum and maximum values in every 10 simulations (for each month, there are 10 in each period).

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

403

between cases and models. Despite these differences, a large increase in nitrogen loadings and less impact on phosphorus loadings were suggested in both studies. Generally, analysis of the GWLF model results showed that the annual total nitrogen loadings were highly sensitive to land use specific curve numbers, nitrogen concentration in runoff and nitrogen accumulation rate (see Appendix B). The annual total phosphorus loadings were predominantly sensitive to land use-specific curve numbers, phosphorus concentration in sediment and sediment delivery ratio. For the five lake catchments, only slight differences existed among those land use-related parameters. Given that the land use area varies between the cases, the land uses contributed to different extents to the total nutrient loadings in the specific cases. The mean values in the deterministic simulation of the base case in the control period (using best estimates of parameter values, but excluding uncertainty) were 1.28 t and 0.12 t for nitrogen and phosphorus, respectively. This is similar to the average of modelled mean results for the control period for the 10 years using the stochastic mode of GWLF model of 1.36 t and 0.11 t for nitrogen and phosphorus, respectively (see Appendix B for details). Furthermore, from the uncertainty analysis, the CV for annual total nutrient loadings was estimated to be on average of 0.27 and 0.46 for nitrogen and phosphorus, respectively, based on 100 realisations for the period 2000–2009 (see Appendix B). We deem this level of model performance sufficient for the first-hand assessments of temperature and precipitation effects on nutrient loadings from climate scenarios in this study.

C A1 (filled marker) are shown in Fig. 8. The nutrients transported by stormwater, surface runoff and groundwater were all taken into account in order to get a seasonal pattern for the total loadings. For nitrogen (Fig. 8a), there was a dramatic increase in loading during the period December–May, which was mainly due to a potentially large increase in groundwater transport. For all months of the year, the integrated nitrogen loading was predicted to be higher in the future period than in the control period. However, due to the large increase in winter–spring, the peak of monthly nitrogen loadings moved from January and July in the control period to December–May in the future scenario. For phosphorus (Fig. 8b), the predicted change in phosphorus loading within individual months from the control period to the future period varied within a limited extent from increasing to decreasing (including negligible), resulting in close to zero effect for a full year (see Fig. 7). The results obtained for the other four lake catchments were similar to those for Råcksta Träsk (not shown). The changes in the seasonal distribution of nitrogen loadings revealed in our modelling results agree with the findings by Moore et al. (2008). They investigated nutrient loadings for the Lake Mälaren catchment (850,800 ha; regional scale), Sweden, using the GWLF model and found that altered seasonal distribution of nutrient loadings was the most profound change for future scenarios (2071–2100). The nutrient loadings increased from early winter, as also suggested in our study. However, in contrast to Moore et al. (2008), we did not find profound seasonal changes in the phosphorus loadings in our future scenario period (2021–2030).

4.4. Monthly nutrient loadings

4.5. Individual impacts from changes in temperature and precipitation

Modelled integrated monthly median loadings for Råcksta Träsk for the control period (open marker) and the most extreme future scenario

In the combined SFA–GWLF model, the annual and monthly nutrient loadings reported above are driven by the combination of temperature

Fig. 9. Total nutrient loadings from 'stormwater + surface runoff' and groundwater to Råcksta Träsk as a function of precipitation (a and c) and (b and d) temperature. (a) and (b) are for nitrogen and (c) and (d) are for phosphorus. The weather data produced are based on a 15% change step for the dataset of daily temperature and precipitation (observed, filled markers) in 2011, with source strengths kept constant. In 2011 mean daily temperature was 8.5 °C and mean annual precipitation was 47.88 cm.

404

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

and precipitation and source strengths with different land use types. In this study, the source strengths and land use were kept constant between the control period and the future scenario period, but the magnitude of changes in temperature and precipitation varied both within seasons and between climate scenarios. In order to further examine the correlation between nutrient loadings and temperature and precipitation individually, sensitivity analyses of nutrient loadings to changes in either temperature or precipitation, under otherwise constant conditions, were carried out for the Råcksta Träsk catchment for 2011 (Fig. 9). For this analysis, 2011 was selected as a reference situation for which model results were produced using observed weather data as input. For the sensitivity analysis, results were generated by simulations with synthetic weather data that contained a 15% change step for the dataset of daily temperature and precipitation (observed) in 2011. Fig. 9a and b shows the results for nitrogen and Fig. 9c and d, the results for phosphorus. The loadings by stormwater and surface runoff and the loadings by groundwater were separated in Fig. 9. For both nitrogen and phosphorus, the loading by stormwater and surface runoff increased with increasing precipitation over the range of precipitation studied (see solid back lines in Fig. 9a and c) while the changes in the loadings were generally less sensitive to changes in temperature (see solid grey lines in Fig. 9b and d). Interestingly, for nitrogen and phosphorus, the loadings by groundwater (dotted lines in Fig. 9) were less sensitive than loadings by surface runoff and stormwater (solid lines in Fig. 9) to both precipitation and temperature over the ranges studied. However, with the projected climate data, groundwater transport of nitrogen showed sensitivity to future climate changes (see Fig. 7). Furthermore, changes in precipitation seemingly have a greater effect on nutrient loadings than changes in temperature. The results in Fig. 9a and c merely reflected a change in volume of precipitation, while the results in Fig. 7 with the projected climate data in addition reflected a change in precipitation events and distributions over the year. Therefore, the projected change in precipitation events and distribution over the year is seemingly more important for the change of nitrogen loadings in the projected future, than the actual change in the annual precipitation volume or in temperature. 4.6. Implications and final discussion The model results presented here indicate that annual nutrient loadings and their seasonal distribution will be marginally (phosphorous) to modestly (nitrogen) affected by climate change effects on precipitation and temperature as projected in our future scenarios. However, results also indicate that the annual groundwater loading of nitrogen may increase dramatically and may be considerably redistributed over the seasons. Such changes can threaten aquatic environments by worsening the eutrophication status of water recipients and affecting the time of algal bloom. In addition, the results suggest potentially different magnitudes of impacts on nitrogen and phosphorus loadings due to climate change, which could cause changes in the limiting nutrient in the selected lakes due to changes in total nitrogen:total phosphorus ratio (Smith, 1982). Taking Råcksta Träsk as an example, during the period 2000– 2009 the total nitrogen:total phosphorus ratio was b6 (Stockholms Stad, 2014) and thus the limiting nutrient was defined as nitrogen, following the criteria by (Smith, 1982). With the predicted trend in total nutrient loadings by stormwater, surface runoff and groundwater, the total nitrogen:total phosphorus ratio in the loading would be N10 for Råcksta Träsk. As a result, over time the limiting nutrient may become phosphorus, which could lead to changes in the structure of aquatic food webs, the seasonal development of phytoplankton and the growth of aquatic macrophytes (George et al., 2010). From a nutrient flow management perspective, vehicular traffic and background atmospheric concentration were identified as two major nutrient sources for the study catchments. Regarding source control, attention may need to be paid to vehicular traffic. Regarding background atmospheric concentration, this may call for further study or monitoring

practices in order to understand the causes and take proper actions. In terms of pathways, there were indications that nitrogen can largely come from both developed areas and natural areas and large amounts of phosphorus are generally from natural areas in the control period. In addition, the major pathways of nitrogen and phosphorus from specific land uses can be identified by the modelled results. Thus, targeting these major pathways may be an efficient way of managing nutrient loadings, with particular treatment of streamflow and land cover management for specific land uses. Furthermore, potentially increased groundwater transport of nitrogen and connected seasonal changes in nitrogen loadings (particularly from December to May) with climate change were indicated. There, thus, appears to be a need for greater understanding of e.g., how to develop nutrient-flow related indicators and how to monitor nutrient flows properly (temporally and spatially). The proposed method of coupling SFA with GWLF can help analyse nutrient flows for management purposes in the future, through quantitatively representing the complex interactions between society and aquatic environments in terms of identifying primary sources, source areas, pathways and loadings. Such an approach may also aid the development of predictive quantitative nutrient flow accounting at urban catchment level, whereby anthropogenic/source-based scenarios in parallel with future climate change can be investigated. It should be noted that nutrient loadings are also sensitive to rainfall event characteristics (e.g., intensity and duration) and extreme events (e.g., flooding) (Pitt and Voorhees, 2004), and the combined SFA-GWLF model used here was unable to capture these changes using daily climate data. Other simplifying assumptions used here include uniform distribution of nutrient sources over the year, no explicit handling of biogeochemistry and no consideration of the effects of climate change on source strengths. Such assumptions were also made in many previous similar studies, e.g., those by Darracq et al. (2005), Wallin (2005) and Moore et al. (2008). Nevertheless, our model results show reasonable agreement with independent estimates of nutrient loadings for the present situation; thus, we propose that our coupled SFA–GWLF model and the obtained model results help to understand potential nutrient loading scenarios with future climate change in urban catchments, in a first approximation. We furthermore note that the common lack of streamflow data for urban catchments currently limit the accuracy of nutrient loading accounting and prediction, thereby motivating further field monitoring of urban water and nutrient flows, particularly associated with groundwater discharge.

5. Conclusions This study quantitatively investigated nutrient diffuse sources and pathways and their loadings to streamflow in urban catchments under the effects of predicted climate change. This was done by coupling source quantifications through SFA (Substance Flow Analysis) with a GWLF (Generalized Watershed Loading Functions) model. Five urban lake catchments, Råcksta Träsk, Judarn, Trekanten, Långsjön and Laduviken in Stockholm, Sweden, were selected as case studies, and potential effects from climate change were explored in a 20-year perspective. In this study, we did not attempt to predict future conditions, but aimed to address potential effects of changes in temperature and precipitation upon climate change on nutrient loadings. While simulated nutrient loadings by surface runoff and stormwater using the combined SFA–GWLF model were deemed to be reasonably certain, the model could only provide upper estimates of nutrient loadings by groundwater for both the control period and future scenarios, due to lack of streamflow data (particularly groundwater discharge) for model calibration in the selected cases. However, model results led us to conclude that the coupled SFA–GWLF model is worthy of further testing and can be used for first-hand investigations of future nutrient loadings in small urban catchments. Moreover, the study highlighted the need for further modelling and field monitoring. Particularly, streamflow data as well as

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

field monitored nutrient flows with different water pathways are necessary for improved accountings and projections of future scenarios. For the five cases studied here, the dominant diffuse sources of nutrients in urban streamflow were indicated to be background atmospheric concentration and vehicular traffic. In addition, from a source area perspective, the major pathways of the nitrogen loading were suggested to be from both developed areas and natural areas in the control period, while phosphorus was largely transported through surface runoff from natural areas. Model results from the coupled SFA–GWLF model using climate scenarios of future temperature and precipitation (2021–2030), neglecting changes in other conditions, indicated increased integrated yearly median nitrogen loadings in all scenarios compared with the control period (2000–20009) for all cases studied. Groundwater was suggested to be the water pathway of nitrogen transport potentially most sensitive to climate change. Scenario C A1, with the largest increase in yearly median precipitation and the smallest decrease in yearly median temperature, resulted in the largest increase in nitrogen loading. The model results also indicated that climate change may potentially cause profound seasonal changes in nitrogen loadings, with a substantial increase from early winter to early summer. The loadings of phosphorus by different water pathways were suggested to be less sensitive to climate change, and the yearly median phosphorus loading and its distribution between pathways and months showed only small changes between the control and future periods. These results are just a small first step in addressing the potential climate change effects of nutrient loadings to urban catchments. However, the potential changes indicated, in particular in nitrogen loading and its redistribution between water pathways and over months, may affect the nutrient status of urban lakes and how nutrient flows need to be managed. Thus further investigations, potentially also including changes in source strengths as a result of climate change and urban development, are urgently needed. An SFA-based modelling strategy such as that used here can help develop predictive quantitative nutrient accounting, e.g., through exploring source-orientated scenarios in parallel with future climate change scenarios for nutrient flow management at urban catchment level. Acknowledgements We express our gratitude to Stina Thörnelöf and Dr. Tonie Wickman, Environment and Health Administration, Stockholm, Sweden, for providing case study data and for valuable discussions. We are also grateful to Dr. Lars Gidhagen, coordinator of SUDPLAN at SMHI, for providing climate scenario data and to Bongghi Hong and Dennis Swaney for providing the latest version of the GWLFXL 2.1.1 model and sharing their expertise. Jiechen Wu also acknowledges financial support from China Scholarship Council. We gratefully acknowledge the helpful comments from the editor, Prof. Damià Barceló Cullerés, and four anonymous reviewers in the reviewing process. Appendices A and B. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.scitotenv.2015.02.041. References Amy, G., 1974. Water Quality Management Planning for Urban Runoff. US Environmental Protection Agency. Barles, S., 2007. Feeding the city: food consumption and flow of nitrogen, Paris, 1801–1914. Sci. Total Environ. 375, 48–58. Bergström, S., Graham, L., 1998. On the scale problem in hydrological modelling. J. Hydrol. 211, 253–265. Bettez, N.D., Marino, R., Howarth, R.W., Davidson, E.A., 2013. Roads as nitrogen deposition hot spots. Biogeochemistry 114, 149–163. Bi, J., Chen, Q., Zhang, L., Yuan, Z., 2013. Quantifying phosphorus flow pathways through socioeconomic systems at the county level in China. J. Ind. Ecol. 17, 452–460.

405

Bouraoui, F., Grizzetti, B., Granlund, K., Rekolainen, S., Bidoglio, G., 2004. Impact of climate change on the water cycle and nutrient losses in a Finnish catchment. Clim. Chang. 66, 109–126. Carey, R.O., Hochmuth, G.J., Martinez, C.J., Boyer, T.H., Dukes, M.D., Toor, G.S., et al., 2013. Evaluating nutrient impacts in urban watersheds: Challenges and research opportunities. Environmental Pollution 173, 138–149. Cui, Q., Brandt, N., Sinha, R., Malmström, M.E., 2010. Copper content in lake sediments as a tracer of urban emissions: evaluation through a source–transport–storage model. Sci. Total Environ. 408, 2714–2725. Dai, T., Wetzel, R.L., Christensen, T.R., Lewis, E.A., 2000. BasinSim 1.0: a windows-based watershed modeling package. Special Report in Applied Marine Science and Ocean Engineering, p. 362. Darracq, A., Greffe, F., Hannerz, F., Destouni, G., Cvetkovic, V., 2005. Nutrient transport scenarios in a changing Stockholm and Mälaren valley region. Water Sci. Technol. 51, 31–38. De Paola, F., Giugni, M., Topa, M.E., Coly, A., Yeshitela, K., Kombe, W., et al., 2013. Intensity– duration–frequency (IDF) rainfall curves, for data series and climate projection in African cities. Poster CLUVA Project 2013. De Risi, R., Jalayer, F., De Paola, F., Iervolino, I., Giugni, M., Topa, M., et al., 2013. Flood risk assessment for informal settlements. Nat. Hazards 69, 1003–1032. Elshkaki, A., van der Voet, E., Timmermans, V., Van Holderbeke, M., 2005. Dynamic stock modelling: a method for the identification and estimation of future waste streams and emissions based on past production and product stock characteristics. Energy 30, 1353–1363. George, G., Järvinen, M., Nõges, T., Blenckner, T., Moore, K., 2010. The impact of the changing climate on the supply and recycling of nitrate. The Impact of Climate Change on European Lakes. Springer, pp. 161–178. Haith, D.A., Mandel, R., Wu, R.S., 1992. GWLF: Generalized Watershed Loading Functions, Version 2.0, User's Manual. Dept. of Agricultural & Biological Engineering, Cornell University, Ithaca, NY. Hong, B., Swaney, D., 2007. GWLFXL 1.2.1 User's Manual. Cornell University, Ithaca, New York. Jalayer, F., De Risi, R., De Paola, F., Giugni, M., Manfredi, G., Gasparini, P., et al., 2014. Probabilistic GIS-based method for delineation of urban flooding risk hotspots. Nat. Hazards 1–27. Johansson, C., Eneroth, K., 2007. TESS Traffic Emissions, Socioeconomic Valuation and Socioeconomic Measures Part 1: Emissions and Exposure of Particles and NOx in Greater. Stockholm Environment and Health Protection Administration, Stockholm. Karlsson, S., Torstensson, G., 2003. Strängkompostering av hästgödsel. JTI-rapportp. 313 (Swedish). Kaye, J.P., Groffman, P.M., Grimm, N.B., Baker, L.A., Pouyat, R.V., 2006. A distinct urban biogeochemistry? Trends Ecol. Evol. 21, 192–199. Kjellström, E., Bärring, L., Gollvik, S., Hansson, U., Jones, C., 2005. A 140-year simulation of European climate with the new version of the Rossby Centre regional atmospheric climate model (RCA3). SMHI Reports Meteorology and Climatologyp. 108. Larm, T., 2000. Watershed-Based Design of Stormwater Treatment Facilities: Model Development and Applications. (Dissertation). KTH Royal Institute of Technology, Stockholm, Sweden. Larm, T., Holmgren, A., 1999. Föroreningsbelastning till sjön Trekanten. Utvärdering av beräkningsmodell för dagvatten. Report No.: nr 44/99. Stockholm Vatten, Stockholm, Sweden, Stockholm (Swedish). Lundgren, J., Pettersson, E., 2004. Practical, environmental and economic evaluation of different options for horse manure management. 16th International Congress of Chemical and Process Engineering, Prague, Czech Republic. Malmqvist, P.-A., 1983. Urban Stormwater Pollutant Sources. Chalmers University of Technology, Dept. of Sanitary Engineering, Goteborg. Miljöförvaltningen, 2002. Beskrivning av problembilden för halterna av kvävedioxid (NO2) och inandningsbara partiklar (PM10) i Stockholms län i förhållande till miljökvalitetsnormerna. Report No.: 5:2002. Stockholms Stad, (Stockholm, Sweden, Swedish). Miljöförvaltningen, 2003. Kartläggning av partikelhalter (PM10) i Stockholms och Uppsala län. Report No.: LVF 2003:1. Stockholms och Uppsala Läns Luftvårdsförbund, (Stockholm, Sweden, Swedish). Moore, K., Pierson, D., Pettersson, K., Schneiderman, E., Samuelsson, P., 2008. Effects of warmer world scenarios on hydrologic inputs to Lake Mälaren, Sweden and implications for nutrient loads. Hydrobiologia 599, 191–199. Mourad, D., Van der Perk, M., 2004. Modelling nutrient fluxes from diffuse and point emissions to river loads: the Estonian part of the transboundary Lake Peipsi/Chudskoe drainage basin (Russia/Estonia/Latvia). Water Sci. Technol. 49, 21. Nakicenovic, N., Alcamo, J., Davis, G., De Vries, B., Fenhann, J., Gaffin, S., et al., 2000. IPCC Special Report on Emissions Scenarios (SRES). Intergovernmental Panel on Climate Change, Geneva. Nash, J., Sutcliffe, J., 1970. River flow forecasting through conceptual models part I—a discussion of principles. J. Hydrol. 10, 282–290. Ogrosky, H.O., Mockus, V., 1964. Hydrology of agricultural lands. Handbook of Applied Hydrology. McGraw-Hill, New York. Pitt, R., Voorhees, J., 2004. WinSLAMM and low impact development. Putting the LID on Stormwater Management, College Park, MD. Schneiderman, E., Järvinen, M., Jennings, E., May, L., Moore, K., Naden, P.S., 2010. Modeling the effects of climate change on catchment hydrology with the GWLF model. The Impact of Climate Change on European Lakes. Springer, pp. 33–50. SMHI, 2014. Stockholm: weather data of temperature and precipitation. [cited 2014 May]. Avaiable from:, www.smhi.se (Internet). Smith, V.H., 1982. The nitrogen and phosphorus dependence of algal biomass in lakes: an empirical and theoretical analysis. Limnol. Oceanogr. 27, 1101–1112.

406

J. Wu, M.E. Malmström / Science of the Total Environment 518–519 (2015) 393–406

Sommer, H., Sieker, H., Jin, Z., Ahlman, S., 2008. Source and flux modelling with STORM-SEWSYS. 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK. Sörme, L., Lagerkvist, R., 2002. Sources of heavy metals in urban wastewater in Stockholm. Sci. Total Environ. 298, 131–145. Stockholms Stad, 2006. Evaluation of the effects of the Stockholm trial on road traffic. Stockholmsförsöket-Evaluation Reports. Stockholmsförsöket, (Stockholm, Sweden). Stockholms Stad, 2007. Adapting to climate change in Stockholm. City of Stockholm Environment and Health Administration, (Stockholm, Sweden). Stockholms Stad, 2014. Nitrogen concentration and phosphorus concentration in lakes in Stockholm. [cited 2014 May]. Available from, http://miljobarometern.stockholm.se/ sub.asp?mp=VP&mo=4&dm=1 (cited 2014 May). Stockholm Vatten, 2000. Vattenprogram för Stockholm 2000. Stockholm Vatten AB, (Stockholm, Sweden, Swedish).

Stockholm Vatten, 2006. Stockholm Water Program 2006–2015. Alfa Print, (Stockholm, Sweden). SUDPLAN, 2014. Stockholm: sustainable urban development planner for climate change adaptation. [cited 2014 May]. Available from:, www.sudplan.cismet.de/run.html (Internet). Wallin, A., 2005. Nutrient Transport Modelling in the Daugava River Basin. Uppsala University. Wu, J., Brandt, N., Malmström, M.E., 2013. Climate change effects on nitrogen loading to urban lakes: the case of Råcksta Träsk, Stockholm, Sweden. 6th International Perspectives on Water Resources & the Environment. The Environmental & Water Resources Institute of the American Society of Civil Engineers (ASCE-EWRI), Izmir Turkey.