Environ Monit Assess (2015) 187: 424 DOI 10.1007/s10661-015-4666-4

Land use/land cover water quality nexus: quantifying anthropogenic influences on surface water quality Cyril O. Wilson

Received: 7 November 2014 / Accepted: 3 June 2015 / Published online: 12 June 2015 # Springer International Publishing Switzerland 2015

Abstract Anthropogenic forces widely influence the composition, configuration, and trend of land use and land cover (LULC) changes with potential implications for surface water quality. These changes have the likelihood of generating non-point source pollution with additional environmental implications for terrestrial and aquatic ecosystems. Monitoring the scope and trajectory of LULC change is pivotal for region-wide planning, tracking the sustainability of natural resources, and meeting the information needs of policy makers. A good comprehension of the dynamics of anthropogenic drivers (proximate and underlying) that influence such changes in LULC is important because any potential adverse change in LULC that may be inimical to sustainable water quality might be addressed at the anthropogenic driver level rather than the LULC change stage. Using a dense time stack of Landsat-5 Thematic Mapper images, a hydrologic water quality and socio-geospatial modeling framework, this study quantifies the role of anthropogenic drivers of LULC change on total suspended solids and total phosphorus concentrations in the Lower Chippewa River Watershed, Wisconsin, at three time steps—1990, 2000, and 2010. Results of the study demonstrated that proximate drivers of LULC change accounted for between 32 and 59 % of the concentration and spatial distribution of total suspended solids, while the extent C. O. Wilson (*) Department of Geography and Anthropology, University of Wisconsin-Eau Claire, 105 Garfield Ave., Eau Claire, WI 54702, USA e-mail: [email protected]

of phosphorus impairment attributed to the proximate drivers ranged between 31 and 42 %. Keywords Land use/land cover change . Anthropogenic drivers . Surface water quality . Chippewa river watershed . Wisconsin

Introduction Land use and land cover (LULC) influence numerous elements within terrestrial and aquatic ecosystems at diverse spatiotemporal scales. Alterations in LULC can alter biodiversity, actual and primary productivity, runoff, and other attributes of ecosystems (Steffens et al. 2004). Anthropogenic factors are mainly responsible for changes in LULC (Meyer and Turner 1992; Lambin et al. 2001). Anthropogenic drivers of LULC change are categorized into proximate and underlying forces (Lambin et al. 2001, 2006). Proximate factors are human activities that operate at the local level such as the individual farmer, household, or community (Lambin et al. 2003). They are regarded as direct drivers of LULC change and can trigger other ecosystem changes that have a relationship with LULC. Underlying drivers of LULC change operate at a much larger scale; they encompass social, political, economic, technological, and other policy-related frameworks (Veldkamp and Fresco 1997; Lambin et al. 2003; Mather 2006b). Proximate and underlying forces of LULC change have the potential to affect surface water quality at diverse spatiotemporal scales. Changes in LULC have been

424 Page 2 of 23

found to affect the pattern of runoff, non-point source pollution production, transportation, and water quality in streams, lakes, and other aquatic environments (Wilson and Weng 2010; Brion et al. 2011; Somura et al. 2012; Tu 2013). Zampella et al. (2007) revealed that 10 % or more of land use alteration can result in significant changes in surface water quality. Heavily polluted water can be detrimental to organisms including humans. Therefore, a good understanding of the dynamics of anthropogenic factors that influence changes in LULC is crucial for urban and regional planning, environmental management, and a host of issues that are connected to natural resources. A plethora of studies have established associations between LULC and surface water quality at varying space-time dimensions (Tong and Chen 2002; Choi et al. 2003; Ren et al. 2003; Goonetilleke et al. 2005; Coats et al. 2008; Tu 2013). Some of these studies examined how LULC proportion and composition within watershed influence surface water quality (Hunsaker and Levine 1995; He 2003). Other investigations have singled out a particular LULC in analyzing the relationship between the latter and surface water quality. For instance, the spatial extent and other characteristics of crop land within a watershed influences nutrient loading and other water quality parameters (Line et al. 2002; Tong and Chen 2002; Zampella et al. 2007; Liu et al. 2009; Brion et al. 2011; Somura et al. 2012). Farm management practices can also influence surface water quality. In this direction, Schilling and Wolter (2009) noted that the use of nitrogen fertilizers within farms in the Des Moines River watershed resulted in water quality impairment. Zhang et al. (2013) observed that an increase in the spatial extent of orchard coupled with loss of forested land resulted in increased nitrogen and phosphorus loading in a watershed. Crop-based agricultural land is not the only culprit of water quality impairment as grazing lands proximate to streams have been documented to exert high impact on water quality (Brion et al. 2011). Such land has the capacity to generate significant sediment and phosphorus loading (Busteed et al. 2009). These cases have shown that agricultural land exerts significant influence on surface water quality in a watershed. Urban/built-up land also has a major influence on surface water quality impairment (Wang 2001; Huang et al. 2013). Yuan et al. (2001) reported that impervious cover is a key contributor of lead in urban catchments. Several studies have pointed that expansion of urban/

Environ Monit Assess (2015) 187: 424

built-up land results in increased phosphorus in water bodies (Waschbusch et al. 1993; Emmerth and Bayne 1996; Winter and Duthie 2000), the generation of high amounts of total suspended solids (TSS) and ammonia (He 2003; Yusop et al. 2005), and the loading of high levels of nitrogen, fats, oil, and grease (FOG) amid other water quality parameters (Tong and Chen 2002; Liu et al. 2009; Wilson and Weng 2010). Within urban settings, golf courses and residential areas undergoing development have been observed to generate high levels of TSS and total nitrogen (TN) (Line et al. 2002). From these studies, it is clear that the greater the size of urban area within a watershed, the higher the likelihood for increased generation and loading of particular pollutants resulting in water quality impairment. To mitigate the role of extensive urban areas on surface water quality, medium- to high-density development has been prescribed (Goonetilleke et al. 2005; Wilson and Weng 2011). While some researchers have documented that forest provides mitigating effects on impaired surface water quality (Park et al. 2011), others have shown that extensive forest cover within a watershed can result in sediments, TN, and total phosphorus (TP) loading in a watershed (Somura et al. 2012). The relationship between LULC and surface water quality has been given considerable attention which has culminated to categorically associating different typologies of LULC to discrete pollutant types. However, efforts directed at categorizing and quantifying anthropogenic contributions to the various types of surface water constituents found in watersheds that have conspicuous human habitation and activities have been inadequate. Researchers found it convenient to use LULC as proxy for anthropogenic drivers in assessing the latter’s role in water quality impairment. Some have utilized multiple linear regression and similar statistical analysis between LULC and surface water quality (Tong and Chen 2002; Park et al. 2011; Seeboonruang 2012; Huang et al. 2013). While this type of approach can rank LULC according to their impacts on surface water quality impairment, it does not show how this phenomenon varies spatially at the sub-watershed level nor the contributions of proximate and/or underlying drivers towards a particular pollutant type. Quite recently, some studies have employed geographically weighted regression (GWR), to account for local variations of LULC impacts on surface water quality (Tu and Xia 2008; Tu 2011, 2013; Sun et al. 2014). This type of approach is an improvement over traditional models discussed above

Environ Monit Assess (2015) 187: 424

but does not directly account for the role of proximate nor underlying forces of LULC change on surface water quality. In a bid to address this hiatus, Huang et al. (2014) segmented natural and anthropogenic controls on water quality based on a space-time dimension. Their approach was a significant improvement vis-àvis previous frameworks but did not differentiate the influence of proximate from underlying drivers. As a result, the spatiotemporal role of anthropogenic drivers of LULC change on surface water quality remains elusive. A comprehensive understanding of the role of these drivers that influence water quality is more important than LULC because any potential adverse change in LULC that might be contrary to sustainable water quality can be addressed at the anthropogenic driver level rather than the LULC stage. To resolve this, it is pivotal to integrate and characterize the role of these drivers in water quality modeling. Results of this type of exercise may be helpful for sustainable water quality planning and management. The overarching goal of this study is to quantify the role of anthropogenic drivers of LULC change, specifically proximate drivers, on surface water quality impairment by employing a socio-geospatial modeling framework. Specific objectives include (1) to characterize LULC spatial composition and configuration between 1990 and 2011 at three time steps and (2) to gauge the role of proximate drivers of LULC change on surface water quality using TSS and TP concentration as water quality parameters. The rest of the paper is arranged as follows: Section 2 provides a brief description of the study area, data used, and methods employed in the study; Section 3 present and discusses results; while section 4 highlights the conclusions.

Materials and methods Study area The Lower Chippewa River watershed (LCRW) is located in West Central Wisconsin (Fig. 1). The Lower Chippewa River extends 177.3 km draining an area spanning 12,521 km2. The watershed lies between 44° 24′ 33″ to 45° 21′ 9″ north latitude and 90° 24′ 44″ to 92° 19′ 58″ west longitude. Elevation varies between 202 m in the southwest where the river drains into the Mississippi and 549 m in the northeast (Martin 1965). The watershed contains numerous lakes especially in the northwest and north central regions. The Chippewa

Page 3 of 23 424

River is the major river in the basin; its key tributaries include Eau Claire, Eau Galle, Yellow, and Red Cedar Rivers. The Lower Chippewa River empties into the Mississippi in the southwestern part of the basin (Fig. 3). The study area is situated in a continental climatic type with average temperature ranging between 1.2 and 12.8 °C for lows and highs respectively with an annual mean temperature of 7 °C (National Climate Data Center 2014). Mean annual rainfall is 788 mm while snowfall averages 1194 mm yearly. The watershed is mostly covered with agricultural and forested LULC, while grass and shrub, water, and settlement are also prominent. The LCRW is mostly rural dotted with smaller communities with the exception of the Eau Claire-Chippewa Falls urban corridor. Major cities in the area include Eau Claire, Chippewa Falls, and Menomonie with an estimated population of 96,000 (U.S. Census 2010). Total population of the LCRW is roughly 280,000. Predominant economic activities in the area include agriculture, services, and mining (West Central Wisconsin Regional Planning Commission 2010). Data Three sets of Landsat-5 Thematic Mapper (TM) satellite images were used in this study. Landsat-5 TM images are available in six reflective bands at a spatial resolution of 30 m (USGS 2013a). Information on acquisition date, rows, and paths of images used in this study is shown in Table 1. The images were carefully selected to minimize variation in vegetation phenology, soil moisture, and other environmental factors. However, due to excessive cloud cover in one of the scenes for 2010, it was not feasible to use images collected in 2010. Instead, images captured in 2011 were utilized for the 2010 time step of the study. Other geospatial data employed include Google Earth high-resolution images (Google Earth 2013), National Agriculture Imagery Program highresolution orthoimagery (USDA 2013), historical aerial photograph of Wisconsin (Wisconsin DNR 2013a), Soil Survey Geographic (SSURGO) data (NRCS 2013), and digital elevation model (USGS 2013b). Continuous stream flow and monthly grab samples of water quality data that span 1980 to 2010 were obtained from USGS archival station discharge data and Wisconsin Surface Water

424 Page 4 of 23

Environ Monit Assess (2015) 187: 424

Fig. 1 Map of the USA showing Wisconsin and the study area. Note: EC Eau Claire, CF Chippewa Falls, and M Menomonie

Integrated Monitoring System (SWIMS) respectively (USGS 2013c; Wisconsin DNR 2013b). Decennial census data at the block group level for 1990, 2000, and 2010 was obtained from the National Historical Geographic Information System (Minnesota Population Center 2011). Data analysis encompassed the processing of Landsat-5 TM images, construction of a distributed hydrologic water quality model, and the development of spatially explicit models to gauge the role of proximate anthropogenic drivers of LULC change on surface water quality.

Table 1 Landsat-5 images used in study TM 1990

TM 2000

TM 2011

Scene

Path 26 row 28

Path 26 row 28

Path 26 row 28

Date

10 September

5 September

3 August

Scene

Path 26 row 29

Path 26 row 29

Path 26 row 29

Date

9 August

11 July

3 August

Scene

Path 25 row 29

Path 25 row 29

Path 25 row 29

Date

19 September

29 August

11 July

TM Thematic Mapper

Satellite image processing Landsat-5 images were geometrically corrected by image-to-map rectification technique for the 1990 images, followed by image-to-image registration for the subsequent images using a third order polynomial equation and nearest neighbor resampling algorithm (Toutin 2004; Jensen 2005). One hundred and twenty ground control points (GCPs) were used in the rectification process with a total root mean square (RMS) error of less than 0.1 pixels for all images. Following geometric correction, the images were corrected for atmospheric interference using the enhanced image-based dark object subtraction (DOS) method (Chavez 1996; Lu et al. 2002). At the completion of atmospheric correction, the study area was delineated (subsetted) from the satellite image scenes. In extracting LULC information from the remotely sensed images, a comprehensive (hard) classification was employed in two stages (Fig. 2). Stage one utilized linear spectral mixture analysis (LSMA), while in stage two, an expert system/decision tree classifier was employed (Ridd 1995; Lu and Weng 2004; Kahya et al. 2010). LSMA was accompanied by the imagebased endmember selection (Roberts et al. 1998). In

Environ Monit Assess (2015) 187: 424

Page 5 of 23 424

Fig. 2 Image processing flow chart. Note: LSMA linear spectral mixture analysis, ML maximum likelihood, TM Thematic Mapper

TM 1990

TM 2000

TM 2011

Preprocessing of TM images

ML

LSMA Hybrid classification (spectral)

Ancillary data

Ancillary data

Expert system/ensemble decision tree

Classified TM 1990

operationalizing this, a forward principal component rotation was used to transform the Landsat-5 TM images to principal component images. At the end of stage one, the study area was disaggregated into urban/built-up, water, forest, non-forested vegetation, and bare soil. For additional information on LSMA, refer to Ridd (1995) and Lu and Weng (2004). Results of stage one classification were pushed into an expert system/decision tree classifier to compartmentalize the urban/built-up land cover into its respective land uses and also to properly delineate agriculture from non-forest vegetation. An expert system/decision tree applies a rule-based approach to images with the aid of user-defined ancillary data (Kahya et al. 2010). For detailed information on expert system/decision tree classifier, please refer to Wentz et al. (2008) and Kahya et al. (2010). At the end of stage two classification, eight LULC classes were produced which include agriculture, commercial, forest, grass, industrial, institutional, residential, and water (Fig. 4). In assessing the accuracy of image classification, 1000 reference points (ground reference information) were obtained for each image through stratified random sampling technique (Congalton 1991). Selection of ground reference information was aided by historical aerial photograph of Wisconsin, National Agriculture Imagery Program high-resolution orthoimagery, and Google Earth high-resolution images for the 1990, 2000, and 2011 images respectively (Google Earth 2013, USDA 2013, Wisconsin DNR 2013a). Overall accuracy of image classification ranged between 85.1

Classified TM 2000

Classified TM 2011

and 87.9 %. Table 2 shows the producers, users, and overall accuracy for each date of image classified. Hydrologic water quality model construction As a result of the dearth in spatial distribution of water quality data in the watershed, a hydrologic water quality modeling framework was adopted to predict and provide data for the entire watershed. T h e m o d e l e m p l o y e d t h e S o i l a n d Wa t e r Assessment Tool (SWAT) informed by a calibration and validation routine—Sequential Uncertainty Fitting Program (SUFI-2) embedded in SWATCUP. SWAT, a fully distributed model, aids in the evaluation of land management practices on water quantity and quality in river basins over long periods of time (Arnold et al. 1993, 1998). SWAT operates on a daily time-step but can be aggregated to monthly or yearly output. The model has an extensive data input which include LULC, soil, elevation, and climate variables. This study constructed three SWAT models for 1990, 2000, and 2010 respectively. For each model, the watershed was automatically delineated with the aid of a 10-m digital elevation model (DEM). The watershed was delineated into 106 sub-watersheds (sub-basins) and 510 hydrologic response units (HRUs). A HRU is a contiguous area within each sub-basin that has uniform land use, soil, slope, and management characteristic and directly drains into a sub-basin (Neitsch et al. 2004).

424 Page 6 of 23

Environ Monit Assess (2015) 187: 424

Table 2 Producers, users, and overall accuracy for classified images Water

Comm

Forest

Grass

Residential

Inds

Agric

Inst

PA(%)

98

77

92

73

75

81

79

88

UA(%)

95

94

83

95

86

91

95

97

2000

PA(%)

94

76

89

63

78

89

75

67

UA(%)

93

100

82

91

85

79

96

96

2011

PA(%)

98

89

98

71

89

92

72

67

UA(%)

99

97

80

84

94

80

86

100

1990

OO 1990 = 85.1 %, OO 2000 = 86.2 %, OO 2011 = 87.9 % Comm commercial, Inds industrial, Agric agriculture, Inst institutional, PA producer’s accuracy, UA user’s accuracy, OO overall accuracy

Sequential Uncertainty Fitting version 2 (SUFI2) program embedded in SWAT-CUP 2012 was utilized in model calibration and validation (Abbaspour 2014). SWAT-CUP is a standalone program that reads SWAT output files with the option of using a variety of parameterization routines (Rouholahnejad et al. 2012). SUFI-2 fits simulated data to measured data and in the process accounts for most sources of uncertainties which include but are not limited to rainfall, conceptual model, parameters, and measured data (Abbaspour 2014). To evaluate whether a model has been calibrated for flow and pollutant, the program uses two key criteria. The first is an index that provides a measure of the model’s ability to capture uncertainty, labeled the P-factor (Rouholahnejad et al. 2012). It is defined as the percentage of measured data bracketed by the 95 % prediction uncertainty (95PPU). The latter is calculated at the 2.5 and 97 % levels. The value of the P-factor ranges between 0 and 100 %. A P-factor of 1(100 %) is a perfect simulation that fully captures model uncertainty; in other words, it accounts for all the model processes. The second measure is the Rfactor that assesses the quality of model calibration. Its calculation is based on the average thickness of the 95PPU divided by the standard deviation of the measured data. The R-factor is denoted by the following equation:  1 m
∑ i¼1 V s;97:5%  V s;2:5% i R¼ m σobs

ð1Þ

where Vs,97.5% and Vs,2.5% are the upper and lower bounds of the 95PPU for a simulated variable Vs,

σobs is the standard deviation of the observed data, while m is the number of parameters fitted. R-factor values ranges between 0 and infinity. An Rfactor of zero illustrates a perfect fit between simulated and measured data. For additional information on the conceptual underpinning of SUFI-2 program, see Abbaspour et al. (2007) and Rouholahnejad et al. (2012). Stream flow data from five United States Geological Survey (USGS) gauging stations were used in calibrating and validating flow for the 1990 and 2000 SWAT models respectively. Total suspended solids (TSS) and total phosphorus (TP) concentration data were converted from concentration to loads prior to calibration and validation at five SWIMS stream monitoring stations for the 1990 and 2000 SWAT models respectively (Fig. 3). The conversion was performed to arrive at units used by SWAT and SUFI-2 programs. The following equation was employed: Z L¼

tb

C ðt Þ*Qðt Þ

ð2Þ

ta

where L is the Bpollutant load^ (kg/d), that is transported through a specified point of the river during the time interval{ta, tb}, C(t) is the pollutant concentration (mg/ L), and Q(t) represents river discharge in (m3/s) at time t. The stream gauge and water quality stations were carefully selected to obtain a balance between spatial distribution and the availability of complete data that covers the study period. Prior to SWAT model calibration, a global sensitivity analysis was performed on 40 SWAT parameters to ascertain parameters that were highly sensitive and reduce the amount of computation and uncertainty during model calibration (Foglia et al. 2009; Abbaspour 2014). These variables were selected

Environ Monit Assess (2015) 187: 424

Page 7 of 23 424

91°45´W

45°10´N

Fig. 3 Simplified SWAT model for Lower Chippewa River Watershed

based on extensive literature search and the author’s knowledge of the watershed’s physical characteristics. At the end of the sensitivity analysis, 13 parameters were found to be highly sensitive to flow, TSS, and TP (Table 3). Using a 4-year warm-up period, the 1990 SWAT model was calibrated for flow, TSS, and TP using mean monthly stream discharge records and loads of the constituents between 1980 and 1990 (Fig. 3). Objective functions used in evaluating model calibration include the P-factor and R-factor discussed above. Model calibration result was satisfactory

with 68 % of observed data bracketed by the 95PPU for the stream gauge and water quality monitoring stations used while the R-factor was less than 1.5 (Fig. 4). The 2000 SWAT model was validated for flow, TSS, and TP using data that span the period 1990 to 2000. Validation results proved satisfactory for all measuring stations employed in model construction (Fig. 5). Following the final model runs in SWAT after calibration and validation, the simulated TSS and TP loads were converted to concentration to

424 Page 8 of 23

Environ Monit Assess (2015) 187: 424

Table 3 Fitted SWAT calibration parameters and their final range Parameter

Description

Lower bound

Upper bound

CN2

Curve number for soil moisture 2

−0.392

0.012

ALPHA_BF

Baseflow alpha factor

0.346

1.104

GW_DELAY

Ground water delay time

150.983

434.017

GWQMN

Threshold depth of water in shallow aquifer

0.539

1.761

ESCO

Soil evaporation compensation factor

0.329

1.028

SOL_AWC

Soil available water capacity

−0.184

0.149

SMTMP

Snowmelt base temperature

−2.535

3.035

GW_REVAP

Ground water Brevap^ coefficient

0.0279

0.147

SURLAG

Surface runoff lag coefficient

5.642

15.607

OV_N

Manning’s Bn^ value for overland flow

0.136

0.248

CH_K(2)

Effective hydraulic conductivity in main channel

14.439

32.711

PRF

Sediment routing factor in main channel

−0.509

1.908

PPERCO

Phosphorus percolation coefficient

3.468

12.142

(Table 4). A combination of geographically weighted regression (GWR) and semi-parametric GWR models were developed and applied. This combined modeling approach was adopted to forestall multicollinearity which often occurs with explanatory variables in GWR models (Wei and Qi 2012; Ivajnsic et al. 2014; Nakaya et al. 2014). In a GWR, the dependent variable Y at each location (g,r) is a function of the explanatory

facilitate comparison with the general format of gauging station data. Socio-geospatial model development To assess the role of anthropogenic factors on TSS and TP concentration and distribution in the watershed, six proximate drivers of LULC change were engaged 45

40

35

Discharge (m3 s-1 )

30

25 95PPU Observed

20

Simulated 15

10

5

0 1

6

11

16

21

26

31

36

41

46

51

56

61

66

71

76

81

Month

Fig. 4 SWAT model calibration of flow 1/1/1980–31/12/1990, sub-basin 36

86

91

96

101 106 111 116 121 126 131

Environ Monit Assess (2015) 187: 424

Page 9 of 23 424

45

40

35

Discharge (m3 s-1 )

30

25 95PPU Observed

20

Simulated 15

10

5

0 1

6

11

16

21

26

31

36

41

46

51

56

61

66

71

76

81

86

91

96

101 106 111 116 121 126 131

Month

Fig. 5 SWAT model validation of flow 1/1/1990–31/12/2000, sub-basin 36

variables at each specific location (Fotheringham et al. 2002; Krivoruchko 2011). The following equation outlines the configuration of a GWR: Y ðg; rÞ ¼ β0 ðg; rÞ þ β1 ðg; rÞX 1 þ β2 ðg; rÞX 2 þ ……:βη ðg; rÞX η þ εðg;rÞ

ð3Þ

where Y(g,r) is the dependent variable (water quality) at location coordinates g and r (centroid of sub-basin); X1, X2, and Xη represent the independent variables at each location; β0, β1, β2, and βη are parameters to be estimated for every location whose coordinates are given by (g,r); and ε(g,r) is the Gaussian error term or residual at location (g,r). Spatial coordinates of the data points are utilized in calculating inter-point distances which are then inputted into a kernel function to develop weights. Table 4 Independent variables used in the study Housing units Median household income Services industry employees Wholesale and retail trade employees Manufacturing and construction employees Agriculture and related industry employees Data source: U.S. Census Bureau

Weights are heavier on observations that are closer to the calibrated coordinate location (g,r) than those that are farther away. A semi-parametric GWR model combines explanatory variables in a local and global framework (Ivajnsic et al. 2014; Nakaya et al. 2014). Equation 4 denotes the framework of a semi-parametric GWR: Y ðg; rÞ ¼

X

β0 ðg; rÞ þ β 1 ðg; rÞX 1 þ β2 ðg; rÞX 2 X þ …:β η ðg; rÞX η þ β1 X 1 þ βη X η þ εðg;rÞ

ð4Þ

where β1X1 and βηXη are respectively fixed parameters and variables that do not vary spatially. A geographical test of spatial variability is performed to ascertain whether a variable should be considered as local or global. Socio-geospatial modeling was conducted using the GWR 4 software package (Nakaya et al. 2014). A Gaussian model was used to fit each model using an adaptive bi-square kernel. The adaptive bi-square kernel is better than the Gaussian fixed and bi-square kernels because it provides a more reliable goodness of fit and also better overcomes the multicollinearity problems between the estimates and produces white noise

424 Page 10 of 23

Environ Monit Assess (2015) 187: 424

residuals (Chasco et al. 2007). An adaptive bi-square weighting function is depicted in the following relation: (h
2 i2 1  d =h ; if d ij < hi ij i ð5Þ W ij ¼ 0 ; otherwise

Table 5 Land use/land cover change (% change)

where Wij is the weight assigned to data point j to estimate the coefficient for regression point i, dij is the straight line distance between observations i and j, and hi denotes the different bandwidths that dictate the fraction of observations to take into consideration in the estimation of regression at location i. An adaptive bi-square kernel has a bandwidth that varies spatially in relation to the variations in density of observations over space. The optimal bandwidth was arrived at by minimizing the corrected Akaike Information Criteria (AICc) as described in Nakaya et al. (2014). For a comprehensive discussion of GWR and the semi-parametric GWR, see Fotheringham et al. (2002) and Nakaya et al. (2014). Two socio-geospatial models were developed for each time step of the study with TSS and TP serving as dependent variables respectively in each model run while socio-demographic data (proximate drivers) served as independent variables. Each model was tested for statistical significance using the pseudo t test at the 95 % confidence interval (p < 0.05).

Results and discussion Land use/land cover change trajectory, 1990 to 2011 Change analysis of LULC demonstrated an increase in the spatial extent of residential, commercial, industrial, and institutional uses over the entire study period. The spatial extent of residential land grew significantly between 1990 and 2000 (63.5 %) and experienced a slightly larger growth between 2000 and 2011 (Table 5). There was notable growth in industrial land use between 1990–2000 and 2000–2011(Table 5). LULC that experienced continuous decline over the study period encompass forest cover and water. For instance, forest cover loss was more pronounced between 2000 and 2011 compared with the previous period investigated. Interestingly, agricultural LULC marginally increased between 1990 and 2000 but witnessed a noticeable decline in the second phase of the period of investigation. Grass LULC, which reduced in size

LULC

1990–2000

2000–2011

Water

−0.7

−0.4

Forest

−1.0

−9.8

Agriculture

0.5

−4.1

Residential

63.5

68.1

Grass

−20.8

95.6

Commercial

17.8

23.6

Industrial

43.2

41.2

Institutional

7.1

33.0

LULC land use/land cover

between 1990 and 2000, demonstrated notable increase in the second period of the study. Increase in residential land can be attributed to population growth over the study period. For instance, analysis of census results shows that population increased over 10 % for each of the study periods while the number of housing units surged at a higher rate. The growth in commercial, industrial, and institutional LULC can be ascribed to the catering of increased services for the expanding population in the watershed over the study period. The overall loss of agricultural land over the study period can be credited to the huge reduction of people that worked in this industry. Analysis of census results for 1990 and 2010 demonstrated a reduction of 78 % in people who work in agricultural and related industries. Forest cover reduction would have occurred to make room for the expansion of residential and the other LULC that increased over the study period. Most of the expansion of residential, commercial, industrial, and institutional LULC occurred in the Chippewa Valley Metropolitan area and the city of Menomonie (Fig. 6). Human influence on the concentration and spatial distribution of total suspended solids The geographically weighted regression (GWR) result for the spatial distribution of TSS concentration in the LCRW demonstrated that on average, 57 % of model variance is explained by socio-demographic information in 1990. The semi-parametric GWR model result for 2000 revealed that 59 % of model variance is explained by socio-demographic information while the component of model variance explained by the semi-parametric model for 2010 declined (Fig. 7). Socio-geospatial

Environ Monit Assess (2015) 187: 424

Page 11 of 23 424

a

b

c

Fig. 6 Land use/land cover maps for 1990 (a), 2000 (b), and 2011 (c)

modeling predictive trajectory shows a decline in the average coefficient of determination (R2) as the average annual concentration of TSS reduced even though the reduction in R2 is not proportional (Fig. 7). Model results for the spatial distribution of TSS concentration over the 20-year period demonstrated that the proximate drivers of LULC change explained between 32 and 59 % of model variance. Other factors besides the proximate drivers are responsible for the remaining component of the spatial distribution of TSS concentration in the LCRW. The spatial composition, configuration and trajectory of LULC account for a number of water quality impairments (Tong and Chen 2000; Goonetilleke et al. 2005; Wilson and Weng 2010). Since LULC composition and configuration are highly influenced by anthropogenic factors especially in areas with human concentration (Veldkamp and Fresco 1997; Fig. 7 Average annual TSS concentration and model coefficient of determination. Note: 1990 = semi-parametric GWR; 2000 and 2010 are GWR

Lambin et al. 2001, 2003; Mather 2006b), one can estimate that the underlying factors of LULC change are also playing a role in the spatial distribution and concentration of TSS in the LCRW. It is possible that this dynamic is also compounded by natural factors including but not limited to slope, precipitation, t e m p e r a tu r e , a n d o t h e r w a t e r s h e d p h y s i c a l characteristics. Chang and Psaris (2013) illustrated that natural landscape factors can be significant in influencing stream temperature. Similar findings have been reechoed elsewhere with researchers establishing positive relationships between some natural factors particularly slope and a number of water quality impairments (Rothwell et al. 2010; Pratt and Chang 2012; Huang et al. 2014). The GWR model developed to predict the role of anthropogenic factors on the spatial distribution of TSS

14 12

R2 = 0.57

10

mg/L

R2 = 0.59

8

R2 = 0.32

6 4 2 0

1990

2000

Year

2010

424 Page 12 of 23

concentration in the LCRW for 1990 shows relatively high R2 in the central north-south span of the watershed (Fig. 8a). Pseudo t test results for this model illustrate a statistically significant positive relationship (p < 0.05) between the number of agricultural personnel and TSS concentrations in portions of the central north-south span of the watershed (Fig. 8b). This suggests that as the number of people working within the agriculture industry increases, the concentration of TSS in the watershed also rises in 1990. However, this is only statistically significant outside the population centers of the Chippewa Metropolitan area and the Menomonie corridor where a proportionally greater number of persons are engaged in non-agricultural activities. Studies have shown that agricultural land is an important agent of TSS loading and concentration (Busteed et al. 2009; Wilson and Weng 2010; Brion et al. 2011). Furthermore, the higher number of agricultural personnel within an area of relatively limited large-scale commercial farming operations in 1990 as exemplified by the LCRW implies a relatively large number of

Environ Monit Assess (2015) 187: 424

individually owned farms with slightly different farm management practices which most likely might have contributed to large TSS concentrations in those portions of the watershed. Median household income and the number of personnel working within manufacturing and construction industries demonstrated a negative statistically significant relationship with the concentration of TSS in the watershed in 1990. Since manufacturing and construction personnel do not necessarily engage in activities that result in huge TSS loading and concentration, the negative relationship seems plausible. Interestingly, the GWR model for 1990 shows that as median household income increases in the watershed, the concentration of TSS abates. This relationship is statistically significant in three clusters of sub-basins within the LCRW (Fig. 8c). It can be argued that as income increases, undeveloped land within those jurisdictions will be converted to developed spaces which are most likely to be impervious surfaces; the latter generally produce lower TSS loadings compared to agricultural lands, bare soil, and other related

Fig. 8 GWR local R2 for TSS and t values for statistically significant socioeconomic variables in 1990. Note: t values significant at p < 0.05; HH household, M & C manufacturing and construction

Environ Monit Assess (2015) 187: 424

underdeveloped land prominent in rural areas. Research has shown that an increase in gross domestic product (GDP) per capita and by extension income results in urbanization and the development of impervious surfaces (Long et al. 2008; Seto et al. 2011). This finding based on the pseudo t test parallels that reported by Siakeu et al. (2004) in their investigation of the spatial dynamics of suspended sediment concentration in central Japan amid land use and demographic changes; they observed that a reduction in agricultural land area due to urbanization resulted in decreased suspended

Page 13 of 23 424

sediment concentration in three quarters of the watersheds. A semi-parametric GWR model was developed to evaluate the role of anthropogenic factors on TSS concentration in the watershed for 2000. The model exhibited higher R2 in the central north-south portion of the watershed which to some extent mirrors that of 1990. Pseudo t test results reveal a statistically significant positive relationship (p < 0.05) between the number of personnel working within the agricultural industry and the concentration of TSS in the watershed (Fig. 9b). Compared to the 1990 pseudo t test result, slightly

Fig. 9 Semi-parametric GWR local R2 for TSS and t values for statistically significant socioeconomic variables in 2000. Note: t values significant at p < 0.05

424 Page 14 of 23

higher number of sub-basins demonstrated this statistically significant positive relationship in 2000. This might have resulted from the marginal increase in agricultural land between 1990 and 2000 (Table 5) and also a probable change in farming culture. Housing units, a local anthropogenic variable, illustrate a statistically significant negative relationship with the concentration of TSS in three portions of the watershed (Fig. 9c). The pseudo t test suggests that as the number of housing units increases in the watershed, the level of TSS concentration decreases. Housing density was temporarily substituted for housing unit during model calibration

Environ Monit Assess (2015) 187: 424

and the result was very similar. This clearly shows that once a rural area is developed through housing and its ancillary infrastructure, the amount of exposed surface will be reduced, thereby curbing the extent of TSS that is generated in those portions of the watershed. Studies have shown that high-density residential areas and development result in reduced footprints for the generation of some non-source point pollutants in watersheds, thereby resulting in relatively lower water quality impairments (Siakeu et al. 2004; Goonetilleke et al. 2005; Wilson and Weng 2011).

Fig. 10 Semi-parametric GWR local R2 for TSS and t values for statistically significant socioeconomic variables in 2010. Note: t values significant at p < 0.05; HH household

Environ Monit Assess (2015) 187: 424

Page 15 of 23 424

A semi-parametric GWR model was developed to predict the relationship between anthropogenic factors and TSS concentration and distribution in the watershed for 2010. The model shows relatively high R2 in the eastern half of the watershed (Fig. 10a). During model calibration, the number of people working within the agricultural sector was not found to be an appropriate local anthropogenic variable; therefore, it was integrated as a global variable. Globally, the number of agricultural personnel demonstrated a negative statistically nonsignificant relationship with the concentration of TSS in the watershed. This negative relationship though not statistically significant can be attributed to the drastic reduction of people within the agriculture industry between 2000 and 2010 (−68 %) and the increase of persons within the service industries (+149 %). Moreover during this period, highly mechanized commercial agriculture became pronounced as Huntsinger Farms, a Wisconsin- and Minnesota-based commercial farming operation acquired large farmlands within the LCRW from initial small holder farmers and other avenues, further contributing to the reduction of individual farmers and the number of farm personnel. The change in spatial relationship between the number of farm personnel and TSS concentration between 2000 and 2010 points to the role of scale change in proximate drivers in modifying the concentration and spatial distribution of non-point source pollutants. Median household income and housing units were found to be statistically significant in explaining the relationship between anthropogenic factors and TSS

Fig. 11 Average annual TP concentration and model coefficient of determination. Note: TP 1990 and TP 2000 = semi-parametric GWR; TP 2010 = GWR

concentration in 2010. Median household income demonstrated a negative statistically significant relationship with TSS concentration in three zones within the watershed (Fig. 10b). Explanation for this pattern mirrors that stated above wherein income normally results in highdensity development thereby reducing room for TSS generation. The anthropogenic variable, housing units this time, records a positive statistically significant relationship with the amount of TSS concentration in portions of the watershed. Probing into the spatial composition and configuration of LULC types in the LCRW as at 2011 to better understand the results of the pseudo t values, it was found that areas which showed positive statistically significant relationship between housing units and TSS concentration are relatively recent developments with low density. Figure 10c shows that these sub-watersheds are to the east and west of the main population center of the Chippewa valley Metropolitan Area. There is a high likelihood that buildings in these new development zones possess very large lots, and other exposed surfaces which will make room for the generation and flushing of TSS, thereby increasing its concentration in the watershed. Studies have shown that low-density residential development results in higher amount of TSS concentration compared with high-density development (Siakeu et al. 2004; Goonetilleke et al. 2005; Wilson and Weng 2011). Land change analysis between 2000 and 2011 demonstrated notable reduction in forest cover especially in the aforementioned sub-watersheds. A number of investigations have demonstrated that forest removal

0.35 R2 = 0.42

0.3

mg/L

0.25 0.2 0.15 R2 = 0.33

0.1 0.05 R2 = 0.31

0 1990

2000

Year

2010

424 Page 16 of 23

results in higher TSS loading and concentrations in water bodies (He 2003; Yusop et al. 2005; Brion et al. 2011; Park et al. 2011). Human influence on the concentration and spatial distribution of total phosphorus The semi-parametric GWR models developed for explaining the spatial distribution of TP concentration in the watershed demonstrated that on average, the proximate drivers accounted for 33 and 31 % of model variance for the 1990 and 2000 models respectively (Fig. 11). The 2010 GWR model illustrated that the

Environ Monit Assess (2015) 187: 424

proximate drivers explained an average of 42 % of model variance in the spatial distribution of TP concentration in the watershed. Figure 11 indicates that GWR and semi-parametric GWR model predictive power increase directly with that of the annual TP concentration in the LCRW. The models developed for the three time steps in this study suggest that underlying and natural drivers account for a greater share of the spatial distribution of TP concentration in the watershed compared with the array of proximate drivers employed. A close examination of the spatial distribution of local R2 for the relationship between proximate drivers and TP concentration in the watershed for 1990 shows

Fig. 12 Semi-parametric GWR local R2 for TP and t values for statistically significant socioeconomic variables in 1990. Note: t values significant at p < 0.05

Environ Monit Assess (2015) 187: 424

that a cluster of southwestern sub-watersheds displayed relatively high coefficient of determination (Fig. 12a). However, sub-basins to the east of the LCRW demonstrated a relatively weak relationship between proximate drivers and the spatial distribution of TP concentration. Pseudo t test values for housing units exhibited a positive statistically significant relationship (p < 0.05) with the concentration of TP in some southwestern subwatersheds and a northeastern sub-watershed (Fig. 12b). This diagnostic suggests that as housing units increase, the generation of TP also rises. Increase in TP concentration in these sub-watersheds cannot only be attributed to agricultural activities but also the application of phosphorus-based fertilizers on lawns especially in low-density residential areas which happen to be prominent in those sub-watersheds highlighted in Fig. 12b. In addition, the wide use of glyphosphate as weed killer in lawns within residential areas of the LCRW might have also contributed to the positive statistically significant t value for housing units. Research has identified that the expansion of urban/built-up land results in increased phosphorus in water bodies (Waschbusch et al. 1993; Emmerth and Bayne 1996; Winter and Duthie 2000). During the 1990 model calibration, the number of agricultural personnel was factored as a global variable as advised by geographical variability test of local coefficients. At the global level, the number of agriculture personnel demonstrated a positive statistically significant relationship (p < 0.05) with the average

Page 17 of 23 424

concentration of TP in the watershed. Moreover, a significant amount of farmsteads were owned by households especially out of the Chippewa Valley Metropolitan Area. The latter would have contributed to the positive association between number of agriculture personnel and TP concentration in the watershed as different farms would have witnessed variations in farm management practices. Farm management culture shows that glyphosphate, a phosphorus-based compound, is extensively used in farms within the LCRW. Most studies that report high amounts of TP concentration in watershed are dominated by agricultural and to some extent lawn management activities (Busteed et al. 2009; Wilson and Weng 2010; Zhang et al. 2013). The pseudo t test for trade personnel exhibited a negative statistically significant relationship with the concentration of TP in three clusters within the LCRW in 1990 (Fig. 12c). This is highly plausible since persons engaged in trade industry do not necessarily embark on activities that will directly result in generating agricultural-related pollutants. The 1990 semiparametric GWR model suggested that as the population of trade personnel increased, the concentration of TP in these sub-watersheds declined. The semi-parametric GWR model developed to assess the influence of proximate drivers on the concentration of TP in the LCRW for 2000 was the least efficacious compared with the other time steps examined in this study. As Fig. 13a demonstrates, only few sub-basins in the east and west of the LCRW explained between 30 and 40 % of the model variance. The only

Fig. 13 Semi-parametric GWR local R2 for TP and t value for a statistically significant socioeconomic variable in 2000. Note: t values significant at p < 0.05

424 Page 18 of 23

proximate driver that displayed statistical significance according to the pseudo t test is housing units which recorded a negative statistically significant relationship with the concentration of TP in two clusters of subwatersheds (Fig. 13b). The negative relationship between the number of housing units in 2000 and the spatial distribution of TP concentration is interesting because in the previous model for 1990, this relationship was positive. A number of factors might have resulted in the negative relationship. The density of housing and its associated development within the sub-basins highlighted in Fig. 13b increased due to population growth between 1990 and 2000 (9.8 %) resulting in proportionally relatively

Environ Monit Assess (2015) 187: 424

smaller lawns and yards. LULC change analysis between 1990 and 2000 for the entire watershed shows significant growth in residential and commercial spaces and reduction in agricultural lands within the aforementioned sub-basins. A triangulation of the pseudo t test result, LULC trajectory, and population change suggest that increase in the intensity of urban development in a mostly rural environment can result in a reduction of the generation of agriculture-related pollutants, in this case phosphorus, within sub-watersheds. Zhao (2008) reported similar findings where the role of agriculture in phosphorus pollution reduced with increased urbanization in a watershed. In a related study, Siakeu et al. (2004) observed a decrease in suspended sediment

Fig. 14 GWR local R2 for TP and t values for statistically significant socioeconomic variables in 2010. Note: t values significant at p < 0.05

Environ Monit Assess (2015) 187: 424

concentration resulting from an increase in urban/builtup area and a corresponding decrease in agricultural land. These studies point to the fact that urbanization can result in the decrease of some pollutants that are typical of non-urban areas. During the construction phase of the model to explain the role of proximate drivers of LULC change on the concentration and spatial distribution of TP in the watershed for 2010, it became feasible to operationalize a fully GWR model where all variables were allowed to locally vary. Local R2 was relatively high in the eastern portion of the watershed (Fig. 14a) while the model underperformed in the northern zone. Interestingly, agricultural personnel illustrated a negative statistically significant relationship with the concentration of TP in some portions of the watershed (Fig. 14b). One would have expected this relationship to be positive in almost all sub-watersheds. A closer examination of LULC, farm culture and management revealed very interesting results. First, the number of people working within the agricultural industry significantly declined between 2000 and 2010 (−68 %) and relatively large-scale mechanized commercial farming took precedence as mentioned in the section BHuman influence on the concentration and spatial distribution of total suspended solids.^ Moreover, as some agricultural lands became converted to urban uses, farm sizes for small holders reduced in area. The LULC change analysis for the entire watershed demonstrated a decline in agricultural land area between 2000 and 2011 (Table 5). All these factors contributed to reduced generation of phosphorus, and therefore, the statistically significant negative relationship exemplified in Fig. 11b. The number of housing units in 2010 also displayed a negative statistically significant relationship with the spatial distribution of TP concentration in clusters of sub-watersheds in the central and eastern portions of the LCRW. When compared with the 2000 LULC map, the 2011 classified image showed conspicuous growth in residential and other urban LULC within these clusters. This development would have reduced surfaces that generate excessive phosphorus and therefore contributed to the negative relationship (Fig. 14c). As Line et al. (2002) reported, phosphorus export was more pronounced in golf courses and pasture areas with very little or no housing units compared to residential areas characterized by significant housing units and other impervious surfaces. In a related investigation, Liu et al. (2009) observed higher influence of

Page 19 of 23 424

agricultural land on phosphorus pollution compared with urban areas in watersheds within Wisconsin. Patch density, a measure of fragmentation/urbanization in a landscape, demonstrated a negative statistically significant relationship with total phosphorus (Sun et al. 2014). This suggests that urbanization reduces total phosphorus concentration in a watershed compared with agricultural land use. However, other researchers have reported otherwise (Soranno et al. 1996; May et al. 1997; Carle et al. 2005; Tu 2011; Tudesque et al. 2014). The majority of these studies were conducted in watersheds dominated by urban/built-up area compared to the mostly rural nature of the LCRW. This study proposes that phosphorus concentration in a watershed that is transitioning from a predominantly rural to urban will reduce as the watershed becomes urbanized. The behavior of phosphorus in an already-urbanized watershed may be different. As mentioned above, the shift in roles of different proximate drivers of LULC change over time will recompartmentalize the influence of each proximate driver and in turn modify the dynamics of water quality in a watershed. In addition, there will be a constant reallocation of influence between proximate, underlying, and natural drivers as the anthropogenic drivers, especially the proximate seem to be in a constant state of flux.

Conclusions This study has demonstrated that the concentration and spatial distribution of TSS and TP in the LCRW is partly influenced by the proximate drivers of LULC change, i.e., number of housing units, median household income, and number of employees by category. This in turn suggests that the underlying and natural forces are also playing a role in the state of water quality. Sociogeospatial modeling illustrated that on average, the proximate drivers accounted for between 32 and 59 % of the concentration and spatial distribution of TSS in the watershed. Moreover, the spatially explicit model suggested that the role of proximate drivers on the spatial distribution and concentration of TSS in the LCRW was more pronounced in 1990 compared with the other time steps investigated in the study. This research has also shown that as a predominantly rural watershed transitions to an urban setting with housing, population, demographic, and infrastructural changes, the concentration of TSS in those sub-basins that

424 Page 20 of 23

experience this change will abate. Furthermore, the intensity of urban development in the LCRW has a direct influence on the reduction of TSS concentration. Suburban-type development characterized by large lots and relatively low population density in the LCRW resulted in comparatively higher TSS concentrations than highdensity development. The study also shows that increase in population and the removal of forest cover can trigger additional TSS loading resulting in higher concentrations in those affected areas. The modeling framework adopted in this study revealed that the proximate drivers accounted for between 31 and 42 % of phosphorus-related water quality impairment in the LCRW. For the time period investigated in this study, underlying and natural forces accounted for a greater share of phosphorusrelated water quality impairments compared with the proximate drivers. Results further suggest that areas with large houses and yards typical of suburban development will exert greater impacts on phosphorus-related water quality problems compared to sub-basins that have higher density development. One mitigating mechanism for phosphorusrelated water quality impairment is an increase in housing development density. The study also shows that phosphorus concentration in a watershed that is transitioning from a predominantly rural to an urban area will reduce as the watershed becomes urbanized while the behavior of phosphorus in an alreadyurbanized watershed may be different. The role of proximate drivers of LULC change on surface water quality is constantly being recompartmentalized over time, thus repeatedly reallocating the influence of underlying and natural factors. It will be interesting to integrate the full array of proximate, underlying, and natural drivers of LULC change in a watershed to quantify the role of each using a space-time analytical framework. The modeling configuration adopted in this study should be applicable to similar watersheds. Acknowledgments This research is partly sponsored by the Office of Research and Sponsored Programs, University of Wisconsin-Eau Claire and through the Simpson Fund, Department of Geography and Anthropology, University of Wisconsin-Eau Claire. The author wishes to thank several agencies including but not limited to the Wisconsin Department of Natural Resources, the West Central Wisconsin Regional Planning Commission, and the GWR 4 Development Team for data, software, and other supporting materials. Finally, the author would like to express gratitude to two anonymous individuals for providing pivotal

Environ Monit Assess (2015) 187: 424 suggestions that helped improve the manuscript and also all those who gave positive feedback on the research during several conference presentations.

References Abbaspour, K. C. (2014). User manual for SWAT-CUP 2012. SWAT calibration and uncertainty programs. [105 pp.]Dubendorf, Switzerland: Ewag: Swiss Fed. Inst. Of Aquat. Sci. and Technol; 2014. Accessed June 5, 2014. http://www.eawag.ch/forschung/siam/software/swat/index Abbaspour, K. C., Yang, J., Maximov, I., Siber, R., Bogner, K., Mieleitner, J., Zobrist, J., & Srinivasan, R. (2007). Modelling hydrology and water quality in the pre-alpine/alpine Thur watershed using SWAT. Journal of Hydrology, 333, 413–430. Arnold, J. G., Allen, P. M., & Bernhardt, G. (1993). A comprehensive surface-ground flow model. Journal of Hydrology, 142, 47–69. Arnold, J. G., Srinivasan, R., Muttiah, R. S., & Williams, J. R. (1998). Large area hydrologic modeling and assessment part 1: model development. Journal of the American Water Resources Association, 34(1), 73–89. Brion, G., Brye, K. R., Haggard, B. E., West, C., & Brahana, J. V. (2011). Land-use effects on water quality of a first-order stream in the Ozark highlands, Mid-southern United States. River Research and Applications, 27, 772–790. Busteed, P. R., Storm, D. E., White, M. J., & Stoodley, S. H. (2009). Using SWAT to target critical source sediment and phosphorus areas in the Wister Lake Basin, USA. American Journal of Environmental Sciences, 5(2), 156–163. Carle, M. V., Halpin, P. N., & Stow, C. A. (2005). Patterns of watershed urbanization and impacts on water quality. Journal of the American Water Resources Association, 41(3), 693–708. Chang, H., & Psaris, M. (2013). Local landscape predictors of maximum stream temperature and thermal sensitivity in the Columbia River Basin, USA. Science of the Total Environment, 461-462, 587–600. Chasco, C., Garcia, I., & Vicens, J. (2007). Modelling spatial variations in household disposable income with geographically weighted regression, Munich personal RePEc arkhive (MPRA), Working Paper, No. 1682. Chavez Jr., P. S. (1996). Image-based atmospheric corrections— revisited and improved. Photogrammetric Engineering and Remote Sensing, 62, 1025–1036. Choi, J., Engel, B., Muthukrishnan, S., & Harbor, J. (2003). GIS based long term hydrologic impact evaluation for watershed urbanization. Journal of the American Water Resources Association, 39(3), 623–635. Coats, R., Larsen, M., Heyvaert, A., Thomas, J., Luck, M., & Reuter, J. (2008). Nutrient and sediment production, watershed characteristics, and land use in the Tahoe basin, California-Nevada. Journal of the American Water Resources Association, 44(3), 754–770. Congalton, R. G. (1991). A review of assessing the accuracy of classification of remotely sensed data. Remote Sensing of the Environment, 37, 35–46. Emmerth, P. P., & Bayne, D. R. (1996). Urban influence on phosphorus and sediment loading of West Point Lake,

Environ Monit Assess (2015) 187: 424 Georgia. Journal of the American Water Resources Association, 32(1), 145–154. Foglia, L., Hill, M. C., Mehl, S. W., & Burlando, P. (2009). Sensitivity analysis, calibration, and testing of a distributed hydrologic model using error-based weighting and one objective function. Water Resources Research, 45(6), 1–18. Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2002). Geographically weighted regression: the analysis of spatially varying relationships. West Sussex:John Wiley. Google Earth. (2013). Google Earth Pro (Version 6.1.0.5001) [Software]. Accessed May 20, 2013 Mountain View, CA: Google Earth. Goonetilleke, A., Thomas, E., Ginn, S., & Gilbert, D. (2005). Understanding the role of land use in urban storm water quality management. Journal of Environmental Management, 74(1), 31–42. He, C. (2003). Integration of geographic information systems and simulation model for watershed management. Environmental Modelling and Software, 18, 809–813. Huang, J., Li, Q., Pontius, R. G., Klemas, V., & Hong, H. (2013). Detecting the dynamic linkage between landscape characteristics and water quality in a subtropical coastal watershed, southeast China. Environmental Management, 51, 32–44. Huang, J., Huang, Y., & Zhang, Z. (2014). Coupled effects of natural and anthropogenic controls on seasonal and spatial variations of river water quality during baseflow in a coastal watershed of Southeast China. PLoS One 9 (3), e91528. doi: 10.1371/journal.pone.0091528 Hunsaker, C. T., & Levine, D. A. (1995). Hierarchical approaches to the study of water quality in rivers. Bioscience, 45(3), 193– 203. Ivajnsic, D., Kaligaric, M., & Ziberna, I. (2014). Geographically weighted regression of the urban heat island of a small city. Applied Geography, 53, 341–353. Jensen, J. R. (2005). Introductory digital image processing: a remote sensing perspective. 3rd ed. Upper Saddle River: Prentice Hall. Kahya, O., Bayram, B., & Reis, S. (2010). Land cover classification with an expert system approach using Landsat ETM Imagery: a case study of Trabzon. Environmental Monitoring and Assessment, 160, 431–438. Krivoruchko, K. (2011). Spatial statistical data analysis for GIS users. Redlands:ESRI Press. Lambin, E. F., Turner, B. L., Geist, H. J., Agbola, S. B., Angelsen, A., Bruce, J. W., Coomes, O. T., Dirzo, R., Fischer, G., Folke, C., George, P. S., Homewood, K., Imbernon, J., Leemans, R., Li, X., Moran, E. F., Mortimore, M., Ramakrishnan, P. S., Richards, J. F., Skanes, H., Steffen, W., Stone, G. D., Svedin, U., Veldkamp, T. A., Vogel, C., & Xu, J. (2001). The causes of land-use and land-cover change: moving beyond the myths. Global Environmental Change, 11, 261–269. Lambin, E. F., Geist, H. J., & Lepers, E. (2003). Dynamic of land use and cover change in tropical and subtropical regions. Annual Review of Environment and Resources, 28, 205–241. Lambin, E. F., Geist, H. J., & Rindfuss, R. R. (2006). Introduction: local processes with global impacts. In E. F. Lambin, & H. J. Geist (Eds.), Land-use and Land cover change: local processes and global impacts (pp. 1–8). New York: IGBP, Springer.

Page 21 of 23 424 Line, D. E., White, N. M., Osmond, D. L., Jennings, G. D., & Mojonnier, C. B. (2002). Pollutant export from various land uses in the Upper Neuse River Basin. Water Environment Research, 74(1), 100–108. Liu, Z., Li, Y., & Li, Z. (2009). Surface water quality and land use in Wisconsin, USA—a GIS approach. Journal of Integrative Environmental Sciences, 6(1), 69–89. Long, H., Wu, X., Wang, W., & Dong, G. (2008). Analysis of urban–rural land-use change during 1995-2006 and its policy dimensional driving forces in Chongqing, China. Sensors, 8, 681–699. Lu, D., & Weng, Q. (2004). Spectral mixture analysis of the urban landscape in Indianapolis with Landsat ETM+ imagery. Photogrammetric Engineering and Remote Sensing, 70(9), 1053–1062. Lu, D., Mausel, P., Brondizio, E., & Moran, E. (2002). Assessment of atmospheric correction methods for Landsat TM data applicable to Amazon Basin LBA Research. International Journal of Remote Sensing, 23(13), 2651–2671. Martin, L. (1965). The physical geography of Wisconsin. Madison: The University of Wisconsin Press. Mather, A. S. (2006b). Driving forces. In H. J. Geist (Ed.), Our earth’s changing land: an encyclopedia of land-use and landcover change, 1, (A-K) (pp. 179–185). West Port: Greenwood Press. May, C. W., Horner, R. R., Karr, J. R., Mar, B. W., & Welch, E. B. (1997). Effects of urbanization on small streams in the Puget Sound lowland ecoregion. Watershed Protection Techniques, 2(4), 483–494. Meyer, W. B., & Turner, B. L. (1992). Human population growth and global land-use/cover change. Annual Review of Ecology and Systematics, 23, 39–61. Minnesota Population Center. (2011). National Historical Geographic Information System: version 2.0. Minneapolis, MN: University of Minnesota. Accessed June 24, 2013 http:// www.nhgis.org Nakaya, T., Charlton, M., Lewis, P., Brunsdon, C., Yao, J. & Fotheringham, S. (2014). Accessed July 6, 2014 https:// geodacenter.asu.edu/drupal_files/gwr/GWR4manual.pdf National Climate data Center. (2014). Land-based station data. Accessed January 11, 2014 http://www.ncdc.noaa.gov/ Neitsch, S. L., Arnold, J. G., Kiniry, J. R., Srinivasan, R., & Williams, J. R. (2004). Soil and water assessment tool input/output file documentation. NRCS [U.S. Department of Agriculture, Natural Resources Conservation Service]. (2013). Soil Survey Geographic (SSURGO) database, Wisconsin. Accessed July 2, 2013. http://glovis.usgs.gov Park, S., Lee, H., Lee, S., Hwang, S., Byeon, M., Joo, G., Jeong, K., Kong, D., & Kim, M. (2011). Relationship between land use and multi-dimensional characteristics of streams and rivers at two different scales. International Journal of Limnology, 47, 107–116. Pratt, B., & Chang, H. (2012). Effects of land cover, topography, and built structure on seasonal water quality at multiple spatial scales. Journal of Hazardous Materials, 209–210, 48–58. Ren, W., Zhong, Y., Meligrana, J., Anderson, B., Watt, W., Chen, J., & Lueng, H. (2003). Urbanization, landuse, and water quality in Shanghai 1947–1996. Environment International, 29(5), 649–659.

424 Page 22 of 23 Ridd, M. K. (1995). Exploring a V-I-S (Vegetation-Impervious Surface-Soil) model for urban ecosystem analysis through remote sensing: comparative anatomy for cities. International Journal of Remote Sensing, 16(12), 2165– 2185. Roberts, D. A., Batista, G. T., Pereira, J. L. G., Waller, E. K., & Nelson, B. W. (1998). Change identification using multitemporal spectral mixture analysis: applications in Eastern Amazonia. In R. S. Lunetta, & C. D. Elvidge (Eds.), Remote sensing change detection: environmental monitoring methods and applications (pp. 137–161). Ann Arbor: Ann Arbor Science Publishers. Rothwell, J. J., Dise, N. B., Taylor, K. G., Allott, T. E. H., Scholefield, P., Davies, H., & Neal, C. (2010). A spatial and seasonal assessment of river water chemistry across North West England. Science of the Total Environment, 408(4), 841–855. Rouholahnejad, E., Abbaspour, K. C., Vejdani, M., Srinivasan, R., Schulin, R., & Lehmann, A. (2012). A parallelization framework for calibration of hydrologic models. Environmental Modelling & Software, 31, 28–36. Schilling, K. E., & Wolter, C. F. (2009). Modeling nitrate-nitrogen load reduction strategies for the Des Moines River, Iowa using SWAT. Environmental Management, 44, 671–682. Seeboonruang, U. (2012). A statistical assessment of the impact of land uses on surface water quality indexes. Journal of Environmental Management, 101, 134–142. Seto, K.C., Fragkias, M., Guneralp, B., & Reilly, M.K. (2011). A meta-analysis of global urban land expansion. PLoS One, 6 (8), e23777. doi:10.1371/journal.pone.0023777 Siakeu, J., Oguchi, T., Aoki, T., Esaki, Y., & Jarvie, H. (2004). Change in riverine suspended sediment concentration in central Japan in response to late 20th century human activities. Catena, 55(2), 231–254. Somura, H., Takeda, I., Arnold, J. G., Mori, Y., Jeong, J., Kannan, N., & Hoffman, D. (2012). Impact of suspended sediment and nutrient loading from land uses against water quality in the Hii River Basin, Japan. Journal of Hydrology, 450-451, 25–35. Soranno, P. A., Hubler, S. L., Carpenter, S. R., & Lathrop, R. C. (1996). Phosphorus loads to surface waters: a simple model to account for spatial pattern of land use. Ecological Applications, 6(3), 865–878. Steffens, W., Sanderson, A., Tyson, P. D., Jager, J., Matson, P. A., Moore, B., Oldfield, F., Richardson, R. J., Schellnhuber, H. J., Turner, B. L., & Wason, R. J. (2004). . Global change and the earth system: a planet under pressure. Berlin:(The IGBP Series), Springer. Sun, Y., Guo, Q., Liu, J., & Wang, R. (2014). Scale effects of spatially varying relationships between urban landscape patterns and water quality. Environmental Management, 54(2), 272–287. Tong, S., & Chen, W. (2002). Modeling the relationship between land use and surface water quality. Journal of Environmental Management, 66(4), 377–393. Toutin, T. (2004). Review article: geometric processing of remote sensing images: models, algorithms and methods. International Journal of Remote Sensing, 25(10), 1893– 1924. Tu, J. (2011). Spatially varying relationships between land use and water quality across an urbanization gradient explored by

Environ Monit Assess (2015) 187: 424 geographically weighted regression. Applied Geography, 31, 376–392. Tu, J. (2013). Spatial variations in the relationships between land use and water quality across an urbanization gradient in the watersheds of northern Georgia, USA. Environmental Management, 51, 1–17. Tu, J., & Xia, Z. G. (2008). Examining spatially varying relationships between land use and water quality using geographically weighted regression I: model design and evaluation. Science of the Total Environment, 407, 358– 378. Tudesque, L., Tisseuil, C., & Lek, S. (2014). Scale-dependent effects of land cover on water physic-chemistry and diatombased metrics in a major river system, the Adour-Garonne basin (South Western France). Science of the Total Environment, 466-467, 47–55. U.S. Census Bureau. (2010). Profile of selected social characteristics: Eau Claire County, WI. Accessed June 12, 2013. http:// factfinder2.census.gov/faces/nav/jsf/pages/index.xhtml USDA [United States Department of Agriculture]. (2013). National Agriculture Imagery Program Accessed June 14, 2013. http://nrcs.usda.gov USGS [United States Geological Survey]. (2013a). Earth Resources Observation and Science Centre. Satellite image collections. Accessed January 10, 2013. http:// glovis.usgs.gov USGS [United States Geological Survey]. (2013b). National Elevation Dataset. Accessed April 18, 2013. http://ned. usgs.gov USGS [United States Geological Survey]. (2013c). USGS Water Data for the Nation. Accessed April 18, 2013. http:// waterdata.usgs.gov Veldkamp, A., & Fresco, L. O. (1997). Reconstructing land use drivers and their spatial scale dependence for Costa Rica (1973 and 1984). Agricultural Systems, 55(1), 19–43. Wang, X. (2001). Integrating water-quality management and landuse planning in a watershed context. Journal of Environmental Management, 61, 25–36. Waschbusch, R. J., Selbig, W. R., & Bannerman, R. T. (1993). Sources of phosphorus in stormwater and street dirt from two urban residential basins in Madison, Wisconsin, 1994-1995. In National conference on tools for urban water resource management and protection. Proceedings February 7-10, 2000 Chicago, IL. Wei, C., & Qi, F. (2012). On the estimation and testing of mixed geographically weighted regression models. Economic Modelling, 29(6), 2615–2620. Wentz, E. A., Nelson, D., Rahman, A., Stefanov, W. L., & Roy, S. S. (2008). Expert system classification of urban land use/land cover for Delhi, India. International Journal of Remote Sensing, 29(15), 4405–4427. West Central Wisconsin Regional Planning Commission (2010). West Central Wisconsin comprehensive plan: 2010-2030. WI:Eau Claire. Wilson, C., & Weng, Q. (2010). Assessing surface water quality and its relations with urban land cover changes in the Lake Calumet Area, Greater Chicago. Environmental Management, 45, 1096–1111. Wilson, C. O., & Weng, Q. (2011). Simulating the impacts of future land use and climate changes on surface water quality in the Des Plaines River Watershed, Chicago Metropolitan

Environ Monit Assess (2015) 187: 424 Statistical Area, Illinois. Science of the Total Environment, 409(20), 4387–4405. Winter, J. G., & Duthie, H. C. (2000). Export coefficient modeling to assess phosphorus loading in an urban watershed. Journal of the American Water Resources Association, 36(5), 1053–1061. Wisconsin Department of Natural Resources. (2013a). Digital orthophoto data. Accessed June 10, 2013. http://dnr.wi.gov Wisconsin Department of Natural Resources. (2013b). Surface Water Integrated Monitoring System (SWIMS). Accessed June 15, 2013. http://dnr.wi.gov/topic/surfacewater/swims Yuan, Y., Hall, K., & Oldham, C. (2001). A preliminary model for predicting heavy metal contaminant loading from an urban catchment. The Science of the Total Environment, 266(1-3), 299–307.

Page 23 of 23 424 Yusop, Z., Tan, L. W., Ujang, Z., Mohamed, M., & Nasir, K. A. (2005). Runoff quality and pollution loadings from a tropical urban catchment. Water Science and Technology, 52(9), 125–132. Zampella, R. A., Procopio, N. A., Lathrop, R. G., & Dow, C. L. (2007). Relationship of land-use/land-cover patterns and surface-water quality in the Mullica River Basin. Journal of the American Water Resources Association, 43(3), 594–604. Zhang, P., Liu, Y., Pan, Y., & Yu, Z. (2013). Land use pattern optimization based on CLUE-S and SWAT models for agricultural non-point source pollution control. Mathematical and Computer Modelling, 58, 588–595. Zhao, J. (2008). Landscape pattern change and its environmental response across multiple spatial scales in tidal plain. East China Normal University, China 107 pp