**3. Methodology**

276 International Perspectives on Global Environmental Change

influenced water nitrogen (Johnson *et al.*, 1997; Fisher *et al.*, 2000; Ahearn *et al.,* 2005), phosphorus (Hill, 1981), total suspended solids (Ahearn *et al.,* 2005) and sediments (Allan *et al.*, 1997). A number of documents have illustrated the increasing urban areas were another significant contributor to the water quality deterioration, since the impervious surface coverage can alter the hydrology and geomorphology of urban streams and give the negative impacts on urban stream ecosystems (Schueler, 1995; Paul and Meyer, 2001; Morse *et al.*, 2003), and runoff from urbanized surfaces carries greater sources of pollutants, which results in the increasing loading of nutrients (Emmerth and Bayne, 1996; Rose, 2002), heavy metals (Norman, 1991; Callender and Rice, 2000), sediment loadings (Wahl *et al.*, 1997) and

In recent years, since 1978 when China has initiated her economic reform and open-door policy, rapid urbanization and economic expansion has resulted in massive land alteration. However, people only focus on the economic growth, and always neglect this factor that economy grows at the expense of the environmental destruction. In this study, therefore, we applied Landsat TM data (2000-2008) to examine the changes of land-use and establish the relationship between land-use types and water quality variables, and give the technical support which can help propose the appropriate strategy that will permit the sustainable regional development and protection of the ecological environment, and understand how it important to assess their potential impacts of landuse types on water quality changes in the watershed scale. This study also demonstrates an example of the issue of how LULC change is linked to water quality, one of the most

Wenyu River watershed is a key area in Beijing (China), belongs to the water systems of the Beiyun River, which is the most intensive area of human activity in Hai River Basin (Figure 1). Wenyu River, the main stream is 47.5 km, which is originates from the south of Yan Mountain and flows from north to south though Haidian, Changping, Shunyi, Chaoyang and Tongzhou Districts, all of these districts are in the core area of Beijing City. Wenyu River is usually called "the mother river" of Beijing, because of all the main streams in Beijing City, it is the only river which originates in the border and never runs

The total area of Wenyu River watershed is 2,478 km2 and the percentage of mountain and flatland area are 40.4% and 59.6%, respectively. The ground elevation in this area is in the region of 15-1000m. And the study area has the terrain characteristics with the high terrain in the northwest and low plain in the southeast. There are many tributaries in this watershed, with the Dongsha, Beisha, and Nansha Rivers in the upper reaches of Wenyu River, meeting in the Shahe Reservoir, and the Lingou, Qing, Ba and Xiaozhong Rivers flowing into the main stream of Wenyu River. The average annual temperature in this watershed is about 11.6 degree Centigrade (for the year 1959-2000). The predominant soil type is cinnamon (53.5%) of the total area. The average annual precipitation is 624.5mm (for the year 1959-2000), more than 80% of a year's total precipitation is concentrated in the flood season from June to September, the average annual water surface evaporation is 1,175mm, and about 42% of a year's evaporation is concentrated from April to June. The average

other contaminants to the near stream waters.

precious resources on earth.

annual runoff is 450 million cubic meters.

**2. Study area** 

dry.

An integrated approach (involving remote sensing, geographic information systems, statistical and spatial analysis, and hydrologic modeling) is used to link the relationship of land use-land cover and water quality in a regional scale. The soft-wares used in this study include ENVI version 4.3, ArcGIS version 9.3, and SPSS version 14.0 for Windows. Figure 2 shows the flowchart of examining the relationship between land-use and water quality.

#### **3.1 Water quality monitoring**

Water samples were collected from twenty-four stations within Wenyu River watershed (see Figure 3) from May to August (on May 22, June 9, July 18 and August 18, respectively) in 2009, and each water sample collection was conducted after the rainfall. Most of these stations distribute in the mid-upper stream area of the Wenyu River watershed.

Assessment of the Impact of Land-Use Types

on the Change of Water Quality in Wenyu River Watershed (Beijing, China) 279

Fig. 2. The flowchart of examining the relationship between land use-land cover and water

quality.

Water quality data are often collected through direct measurement in situ. To some variables cannot be measured in situ, a sample must be taken and then analyzed in a laboratory. In this research, water samples are analyzed to obtain six water quality variables, as Table 1 listed. The variable of DO is in situ measured using Portable Dissolved Oxygen Analyzer, TOC is analyzed in the laboratory using Total Organic Carbon Analyzer, and the other variables are measured according to National standardized water quality detection method (State Environmental Protection Administration of China, 2002).


Table 1. Water Quality parameters selection in this study.

#### **3.2 Sub-watershed delineation**

Because the 24 water sampling points of this study locate across a range of land uses, geology types, and stream orders within the entire Wenyu River watershed. Thus, the subwatersheds within Wenyu River watershed should be firstly delineated, and Arc Hydro Model is employed to do this job. Arc Hydro Model was developed by a consortium for geographic information systems (GIS) in water resources, integrated by the University of Texas' Center for Research in Water Resources (CRWR) and the Environmental Systems Research Institute (ESRI) during the years 1999-2002. The Arc Hydro data model is a conceptualization of surface water systems and describes features such as river networks, watersheds and channels. The data model can be the basis for a "hydrologic information system", which is a synthesis of geospatial and temporal data supporting hydrologic analysis and modeling (Maidment, 2002). The Arc Hydro tools are a set of utilities developed based on the Arc Hydro data model, and operating in the ArcGIS environment. These tools can be used to process a digital elevation model raster (DEM) to delineate subwatersheds.

The major data used to delineate the sub-watersheds is the 30 meter DEM (Digital Elevation Model) data set for China, which is a part of ASTER (Advanced Space-borne Thermal Emission and Reflection Radiometer) Global 30m DEM topographic data set and available for download free of charge from the NASA's Land Process Distributed Active Archive Center, at URL https://wist.echo.nasa.gov/api/. Using Boundary vector of the study area, the DEM for the study area can be obtained. In this process, higher threshold will result in less dense stream network and less internal sub-watersheds; when the value of threshold decrease, a relatively dense stream network and more internal sub-watersheds will be obtained. In this research, the value of 50000 is applied as the threshold value, the resultant stream network and sub-watershed delineation rasters are displayed in Figure 4. It can also be found 42 sub-watersheds are delineated within Wenyu River watershed when 50000 is used as the threshold value.

Assessment of the Impact of Land-Use Types on the Change of Water Quality in Wenyu River Watershed (Beijing, China) 279

278 International Perspectives on Global Environmental Change

Water quality data are often collected through direct measurement in situ. To some variables cannot be measured in situ, a sample must be taken and then analyzed in a laboratory. In this research, water samples are analyzed to obtain six water quality variables, as Table 1 listed. The variable of DO is in situ measured using Portable Dissolved Oxygen Analyzer, TOC is analyzed in the laboratory using Total Organic Carbon Analyzer, and the other variables are measured according to National standardized water quality detection method (State Environmental Protection

Variable Name Chemical Formula or Abbreviation Unit Dissolved Oxygen DO mg/l Chemical Oxygen Demand COD mg/l Total Nitrogen TN mg/l Nitrate NO3- N mg/l Total Phosphorous TP mg/l Phosphate PO4- P mg/l

Because the 24 water sampling points of this study locate across a range of land uses, geology types, and stream orders within the entire Wenyu River watershed. Thus, the subwatersheds within Wenyu River watershed should be firstly delineated, and Arc Hydro Model is employed to do this job. Arc Hydro Model was developed by a consortium for geographic information systems (GIS) in water resources, integrated by the University of Texas' Center for Research in Water Resources (CRWR) and the Environmental Systems Research Institute (ESRI) during the years 1999-2002. The Arc Hydro data model is a conceptualization of surface water systems and describes features such as river networks, watersheds and channels. The data model can be the basis for a "hydrologic information system", which is a synthesis of geospatial and temporal data supporting hydrologic analysis and modeling (Maidment, 2002). The Arc Hydro tools are a set of utilities developed based on the Arc Hydro data model, and operating in the ArcGIS environment. These tools can be used to process a digital elevation model raster (DEM) to delineate sub-

The major data used to delineate the sub-watersheds is the 30 meter DEM (Digital Elevation Model) data set for China, which is a part of ASTER (Advanced Space-borne Thermal Emission and Reflection Radiometer) Global 30m DEM topographic data set and available for download free of charge from the NASA's Land Process Distributed Active Archive Center, at URL https://wist.echo.nasa.gov/api/. Using Boundary vector of the study area, the DEM for the study area can be obtained. In this process, higher threshold will result in less dense stream network and less internal sub-watersheds; when the value of threshold decrease, a relatively dense stream network and more internal sub-watersheds will be obtained. In this research, the value of 50000 is applied as the threshold value, the resultant stream network and sub-watershed delineation rasters are displayed in Figure 4. It can also be found 42 sub-watersheds are delineated within Wenyu River watershed when 50000 is

Administration of China, 2002).

**3.2 Sub-watershed delineation** 

watersheds.

used as the threshold value.

Table 1. Water Quality parameters selection in this study.

Fig. 2. The flowchart of examining the relationship between land use-land cover and water quality.

Assessment of the Impact of Land-Use Types

**3.3 LULC classification in the study area** 

on the Change of Water Quality in Wenyu River Watershed (Beijing, China) 281

To those sub-watersheds containing in-situ measured water quality data, it is very clear about the water quality status there and obtain the mean values of each water quality

Landsat TM data are used to extract the land use-land cover information of the Wenyu River watershed. Landsat TM is appropriate for the purpose in this research because it is free online and can be downloaded easily. Its spatial resolution is 30 meter which will be appropriate to conduct land use analysis of the watershed of Wenyu River. One nearly cloud-free Landsat 5 TM image covering the study area is acquired from the USGS website, http://glovis.usgs.gov. Table 2 describes the general information of this downloaded image.

Landsat Scene Identifier LT51230322009201IKR00 WRS Path/ROW \* 123/032 Data Acquired 2009/07/20 Cloud Cover 3.58% Corner Upper Left 41°16'19"N/ 115°53'07"E Corner Upper Right 40°57'24"N/118°02'53"E Corner Lower Left 39°41'38"N/115°24'26"E Corner Lower Right 39°23'08"N/117°31'21"E \* WRS means The Worldwide Reference System, which is a global notation used in cataloging Landsat

To extract land covers of Wenyu River watershed from Landsat TM 5 data, the supervised classification method is adopted in this research, which is the procedure most frequently used for quantitative analysis of remote sensing data, and the maximum likelihood algorithm is employed to detect the land cover types in ENVI software. Based on the priori knowledge of the study area and additional information from previous research in Wenyu River watershed, a classification system concerned with six land classes has been established for this study area, including forest, farmland, urban, village, bare land and the water

1 Forest land Coniferous & deciduous forest, trees covers, shrubs with

2 Farmland Cropland and pasture, Orchards, other agriculture land

4 Village area Located in the rural areas, surrounding the urban area and has a relatively low population density

areas, uncultivated agricultural lands

5 Bare land Areas with no vegetation cover, stock quarry, stony

6 Water body Seas, lakes, reservoirs, rivers and wetland

Table 3. Land use-land cover classification scheme used in TM data.

Residential, commercial, industrial, transportation, and communications facilities; the area of intensive use with much

of the land covered by structures and high population

density, usually located in the center of a city

data; both Landsat 5, 7 follow the WRS-2, and Landsat 1,2,3,4 follow the WRS-1. Table 2. The general information of downloaded Landsat 5 TM scene.

bodies, the description of these land cover classes are presented in Table 3.

partial grassland

No. Land Cover Type Description

3 Urban area

parameters of these sub-watershed through the statistical computing process.

Fig. 3. Water Quality Sampling Points in Wenyu River Watershed (Landsat TM5 image).

Fig. 4. The sub-watersheds delineation results generated by using the threshold value of 50000.

To those sub-watersheds containing in-situ measured water quality data, it is very clear about the water quality status there and obtain the mean values of each water quality parameters of these sub-watershed through the statistical computing process.

#### **3.3 LULC classification in the study area**

280 International Perspectives on Global Environmental Change

Fig. 3. Water Quality Sampling Points in Wenyu River Watershed (Landsat TM5 image).

Fig. 4. The sub-watersheds delineation results generated by using the threshold value of 50000.

Landsat TM data are used to extract the land use-land cover information of the Wenyu River watershed. Landsat TM is appropriate for the purpose in this research because it is free online and can be downloaded easily. Its spatial resolution is 30 meter which will be appropriate to conduct land use analysis of the watershed of Wenyu River. One nearly cloud-free Landsat 5 TM image covering the study area is acquired from the USGS website, http://glovis.usgs.gov. Table 2 describes the general information of this downloaded image.


\* WRS means The Worldwide Reference System, which is a global notation used in cataloging Landsat data; both Landsat 5, 7 follow the WRS-2, and Landsat 1,2,3,4 follow the WRS-1.

Table 2. The general information of downloaded Landsat 5 TM scene.

To extract land covers of Wenyu River watershed from Landsat TM 5 data, the supervised classification method is adopted in this research, which is the procedure most frequently used for quantitative analysis of remote sensing data, and the maximum likelihood algorithm is employed to detect the land cover types in ENVI software. Based on the priori knowledge of the study area and additional information from previous research in Wenyu River watershed, a classification system concerned with six land classes has been established for this study area, including forest, farmland, urban, village, bare land and the water bodies, the description of these land cover classes are presented in Table 3.


Table 3. Land use-land cover classification scheme used in TM data.

Assessment of the Impact of Land-Use Types

delineated sub-watersheds.

**3.5 An exponential model** 

follows:

use type and *NPSi* .

on the Change of Water Quality in Wenyu River Watershed (Beijing, China) 283

In statistical researches, Spearman's rank correlation coefficient is a non-parametric measure of statistical dependence between two variables, which allows us to easily identify the strength of correlation within a data set of two variables, and whether the correlation is positive or negative. The absolute value of the correlation coefficient, with the range from 0 to 1, indicates the strength, with larger absolute values indicating stronger relationships. The significance level (also termed as p-value) is the probability of obtaining results as extreme as the one observed. If the significance level is very small (p value is less than 0.05), the correlation is significantly raleted at 95% confidence level, and the two variables are linearly related. The data set, which are used in the Spearman's rank correlation process to determine the relationships between land use cover and water quality in this research, includes the land use-land cover variables (%) and the water quality variables (mg/L) of the

Delivery of non-point source pollutants from discrete upstream contributing zones to a particular downstream point is a multi-step, often episodic, process (Phillips, 1989). During the rainfall event, the pollutants released from different land use types will flow through various land covers with the surface runoff, continuing to be absorbed, deposited and released, and eventually enter the nearest stream water. A first-order rate equation can be used for modeling nutrient attenuation in flow through various land uses to the nearest stream (Phillips, 1989). Thus in most cases, the concentration of nutrients or total suspended solids ( *NPSi )* at a sample point received from a basin *i*, can be described in the form of an exponential model (Fetter 1994; Basnyat *et al.,* 1999; Basnyat *et al.,* 2000) as

1 2 3 4 56 ( ) *Forest Farmland Urban Villa i i i ii i ge Bare Water NPS e <sup>i</sup>*

parameters that specify the direction and strength of the relationships between each land

Based on the linkage model, multiple regression models were applied to each of water quality variables: total nitrogen, nitrate, total phosphorous, phosphate, chemical oxygen demand and dissolved oxygen, respectively. A backwards stepping approach is employed to isolate a final model with only significant independent variables included. In Backward approach, all the predictor variables will go into the model firstly. The weakest predictor variable is then removed and the regression re-calculated. If this significantly weakens the model, the predictor variable will re-entered, otherwise it will be deleted. This procedure

The purpose of multiple regression process is to predict a single variable (dependent variable) from one or more independent variables. For each model, the initial fixed independent variables are LULC variables (forest, farmland, urban, village, bare and water). The dependent data of water quality parameters and the independent data of land use variables will be natural log transformed to meet the assumptions of normality, as determined via graphical evaluation of standard diagnostic graphs. Finally, goodness-of-fit of final significant statistical models will be evaluated by scatter plot to compare the

   

1 , <sup>2</sup> , 3 , <sup>4</sup> , 5 and 6 are

(1)

 

Where *NPSi* is the dependent variable, α is the intercept

observed data against equivalent model prediction.

 

will repeated until only useful predictor variables remain in this model.

During the process of supervised classification, the collection of training sites constitutes a very critical stage and it is essential that all the required classification classes are sampled. The quality of a supervised classification depends on the quality of the training sites. In order to select the accurate training sites, different band combinations are used to identify the different land categories, according to Landsat TM Band spectral characteristics. Figure 5 displays the generated land use-land cover map of Wenyu River watershed in 2009.

Fig. 5. Land use-land cover map of Wenyu River watershed in 2009 from Landsat TM 5 data.

The LULC map shows that, upper region of Wenyu River has significantly more forest land with higher elevation, while the middle region of the research watershed has a higher percentage of urban area and the major land types in the lower region are village and farmland. The different regions in Wenyu River watershed differ significantly in terms of percentage of forest, urban, village and farmland covers.

#### **3.4 Spearman's rank correlation**

Since most of the water quality variables do not distribute normally, the statistical analyses are confined to non-parametric statistical tests, spearman's rank correlation analyses are used to explore the relationships between land use types and water quality indicators in Wenyu River Watershed. And this statistical analyses are performed using SPSS 14.0 for Windows.

In statistical researches, Spearman's rank correlation coefficient is a non-parametric measure of statistical dependence between two variables, which allows us to easily identify the strength of correlation within a data set of two variables, and whether the correlation is positive or negative. The absolute value of the correlation coefficient, with the range from 0 to 1, indicates the strength, with larger absolute values indicating stronger relationships. The significance level (also termed as p-value) is the probability of obtaining results as extreme as the one observed. If the significance level is very small (p value is less than 0.05), the correlation is significantly raleted at 95% confidence level, and the two variables are linearly related. The data set, which are used in the Spearman's rank correlation process to determine the relationships between land use cover and water quality in this research, includes the land use-land cover variables (%) and the water quality variables (mg/L) of the delineated sub-watersheds.

#### **3.5 An exponential model**

282 International Perspectives on Global Environmental Change

During the process of supervised classification, the collection of training sites constitutes a very critical stage and it is essential that all the required classification classes are sampled. The quality of a supervised classification depends on the quality of the training sites. In order to select the accurate training sites, different band combinations are used to identify the different land categories, according to Landsat TM Band spectral characteristics. Figure 5

Fig. 5. Land use-land cover map of Wenyu River watershed in 2009 from Landsat TM 5 data. The LULC map shows that, upper region of Wenyu River has significantly more forest land with higher elevation, while the middle region of the research watershed has a higher percentage of urban area and the major land types in the lower region are village and farmland. The different regions in Wenyu River watershed differ significantly in terms of

Since most of the water quality variables do not distribute normally, the statistical analyses are confined to non-parametric statistical tests, spearman's rank correlation analyses are used to explore the relationships between land use types and water quality indicators in Wenyu River Watershed. And this statistical analyses are performed using SPSS 14.0 for

percentage of forest, urban, village and farmland covers.

**3.4 Spearman's rank correlation** 

Windows.

displays the generated land use-land cover map of Wenyu River watershed in 2009.

Delivery of non-point source pollutants from discrete upstream contributing zones to a particular downstream point is a multi-step, often episodic, process (Phillips, 1989). During the rainfall event, the pollutants released from different land use types will flow through various land covers with the surface runoff, continuing to be absorbed, deposited and released, and eventually enter the nearest stream water. A first-order rate equation can be used for modeling nutrient attenuation in flow through various land uses to the nearest stream (Phillips, 1989). Thus in most cases, the concentration of nutrients or total suspended solids ( *NPSi )* at a sample point received from a basin *i*, can be described in the form of an exponential model (Fetter 1994; Basnyat *et al.,* 1999; Basnyat *et al.,* 2000) as follows:

$$\text{NPS}\_{i} = \alpha e^{\{\beta\_{1}\text{Fores}\_{i} + \beta\_{2}\text{Farmland}\_{i} + \beta\_{3}\text{LIrhan}\_{i} + \beta\_{4}\text{Village}\_{i} + \beta\_{5}\text{Rare}\_{i} + \beta\_{6}\text{Vater}\_{i}\}} \tag{1}$$

Where *NPSi* is the dependent variable, α is the intercept 1 , <sup>2</sup> , 3 , <sup>4</sup> , 5 and 6 are parameters that specify the direction and strength of the relationships between each land use type and *NPSi* .

Based on the linkage model, multiple regression models were applied to each of water quality variables: total nitrogen, nitrate, total phosphorous, phosphate, chemical oxygen demand and dissolved oxygen, respectively. A backwards stepping approach is employed to isolate a final model with only significant independent variables included. In Backward approach, all the predictor variables will go into the model firstly. The weakest predictor variable is then removed and the regression re-calculated. If this significantly weakens the model, the predictor variable will re-entered, otherwise it will be deleted. This procedure will repeated until only useful predictor variables remain in this model.

The purpose of multiple regression process is to predict a single variable (dependent variable) from one or more independent variables. For each model, the initial fixed independent variables are LULC variables (forest, farmland, urban, village, bare and water). The dependent data of water quality parameters and the independent data of land use variables will be natural log transformed to meet the assumptions of normality, as determined via graphical evaluation of standard diagnostic graphs. Finally, goodness-of-fit of final significant statistical models will be evaluated by scatter plot to compare the observed data against equivalent model prediction.

Assessment of the Impact of Land-Use Types

**Subwatershed Number** 

**Different Spatial Areas** 

Upstream Mountain Area

Midstream Urban Area

Downstream Plain Area

on the Change of Water Quality in Wenyu River Watershed (Beijing, China) 285

**Water Sampling Sites** 

w2 Sites 5, 6

w4 Sites 1, 2, 3, 4

w27 Site 13

w28 Site 15, 16

w34 Site 8 w35 Site 7

w9 Site 20 w15 Sites 17, 19

w22 Site 18 w31 Site 22 w32 Site 23 w42 Site 24

Table 4. Three different spatial areas definition within Wenyu River Watershed.

concentration, including TN, NO3- N, TP, PO4- P, COD and DO.

**4.2 Water quality comparison between different land-use types** 

Through the statistical computing process, water quality information in Upstream Mountain Area, Midstream Urban Area and Downstream Plain Area can be obtained based on the measured water quality data at total 24 water sampling sites. These water quality statistical information include the mean value (the sum of all observations divided by the number of observations) and the standard error of the mean (SEM, calculated by dividing the standard deviation by the square root of the sample size) of six water quality prameters's

In order to conduct the further analysis of the relationship between land use and the water quality within Wenyu River watershed, in this section, the sub-watersheds are divided into

w33 Sites 9, 10, 11, 12

**Area Characteristics** 

Wenyu River Watershed and with the higher elevation; With the only significant land

quality from human activities.

Wenyu River Watershed; With gently sloping surface; With the notable land use of Urban and village; High density of population; Considerable influence on water quality from human

Wenyu River Watershed; With gently sloping surface; With the dominant land use of production agriculture; Relatively low density of

 Certain influence on water quality from agriculture

Lying in the upstream of

 Sparse human population; Less influence on water

use of forest;

activities.

population;

activities.

w26 Site 14 Lying in the midstream of

w8 Site 21 Lying in the downstream of
