Home Articles Simulating Impacts of Residential Patterns on Hydrology of an Urbanizing Watershed by...

Simulating Impacts of Residential Patterns on Hydrology of an Urbanizing Watershed by Using SWAT and ArcView® GIS

M. Rafee Majid
Department of Urban & Regional Planning
Universiti Teknologi Malaysia
81310 UTM, Skudai
Malaysia
[email protected]

ABSTRACT
Urban land use is the main contributor of impervious surface, the most important factor influencing hydrologic behaviour of urban watersheds. In the quest to better understand the relationship between impervious surface and hydrologic behavior of a watershed, Geographic Information System (GIS) and remote sensing are now commonly employed to parameterize hydrologic models. This paper describes the use of a hydrological model called SWAT (Soil and Water Assessment Tool) to assess the impact of different residential-development scenarios on watershed behaviour. The pre- and post-processing of SWAT’s input/outputs are done in ArcView® while impervious surface is estimated using remote sensing images. The study investigates the relationship between stream flow and runoff ratio as a function of percent impervious cover for a small watershed. A calibrated model is used to model runoff and stream flow within the watershed under eleven different residential-development scenarios, varying in density and design of development. The significant difference between these scenarios in term of their effects on hydrological behavior is the amount of impervious surface. The results indicate that there are differences in the potential runoffs generated by the different scenarios. At the same design capacity, the amount of runoff from compact developments simulated in the model is less than that from low-density developments. The study thus confirms a relationship between runoff and pattern of urban development and demonstrates that watershed model can be used to understand the impact of development characteristics on the hydrology of watersheds. The model can generate “what-if” scenarios of the watersheds under study, something that can be used by land use planners in making decisions on a variety of land-use options.

INTRODUCTION
Land use regulations are one of several tools urban planners use to control physical characteristics of developing landscapes. Through these regulations, planners can impose restrictions on various development variables including land use types and activities, density, open space and even extent of impervious surface. These variables, in turn, influence environmental processes of the atmosphere, surface and subsurface hydrologic systems and even weather and climate regimes. The intrinsic relationships between land use regulations and the physical structure of an urban watershed suggest that there should be mechanisms through which planners can evaluate the environmental consequences of existing regulations and improve the scientific basis of future decision making to further minimize negative effects on the watershed. An integrated application of geographical information system (GIS), remote sensing and numerical modeling has the potential to contribute to this informed decision-making process by providing synoptic data collection capabilities and analysis techniques that can generate pertinent information about watershed implications of planning decisions. While GIS has long been used by planners, its coupling with numerical models and remote sensing to solve more complex problems or to find better explanations for existing problems is relatively recent.

One example of such integrated uses is the hydrological modeling of a watershed. For this, data from GIS and remote sensing can be together used to parameterise the simulation of a hydrological model to evaluate the impact of planning decisions on watersheds. Using the latest technique in remote sensing together with GIS data, an accurate and geographically extensible impervious surface prediction model can be developed. The model can then be used to quantify the amount of impervious surface and the results used to find the relationship between land use/subdivision regulations and impervious area. The information obtained can subsequently be used as an input to a hydrological model to assess the hydrological impact of land use regulations on a watershed.

This paper, however, discusses only the hydrological simulation part of the application. This is part of the study carried out by the author investigating the application of remote sensing and GIS in assessing the influence of density and subdivision design on residential-area imperviousness with a demonstration on the application of the results for hydrological simulation of a watershed. For techniques on estimating impervious surface using remote sensing, readers are encouraged to refer to Majid (2006), Rashed et al. (2003), Yang et al. (2003), Wu and Murray (2002) and Ward et al. (2000). A good review of the techniques is provided by Slonecker et al. (2001).

The objectives of the paper are: 1) to analyse the change in the flow behaviour of an urbanizing watershed due to different densities and subdivision designs; and 2) to demonstrate an application of GIS coupled with numerical modeling in predicting the implications of urban planning (land use) regulations. The findings from this study can help one understand how a combination of density and subdivision design affects the hydrology of a watershed. In addition, this study adds to the application of GIS coupled with numerical modeling to facilitate new avenues of exploratory spatial data analysis.

Impervious Surface, Runoff And Subdivision Design
Impervious surface can generally be defined as any material that prevents the infiltration of water into the soil. Imperviousness, meanwhile, is an extent of impervious surface coverage expressed as a percentage of the whole area being considered. In a subdivision, most prevalent impervious surface is rooftops and roads while other types are driveways, sidewalks and patios. As development alters the natural landscape, the percentage of land covered by impervious surface, or the imperviousness level, increases. Impervious surfaces have not only been identified as a key environmental indicator of urbanization, but have also been acknowledged as major contributors to the environmental impacts of urbanization (Arnold and Gibbons, 1996). As the natural landscape is paved over, a chain of events is initiated that typically ends in degraded water resources. This chain begins with alterations in the hydrologic cycle, the way that water is transported and stored. These changes, depicted in Figure 1.0, have long been understood in hydrology. As impervious coverage increases, the volume and velocity of surface runoff increase with a corresponding decrease in infiltration. The larger volume of runoff and the increased efficiency in water conveyance through pipes and artificial channels leave little chance for infiltration and result in storm flows that are greater in volume and faster in reaching peak flows (Leopold, 1968; Tourbier and Westmacott, 1981). The shift away from infiltration reduces groundwater recharge, lowering water tables. This both threatens water supplies and reduces the groundwater contribution to stream flow, which can result in intermittent or dry stream beds during periods of low flow (Dunne and Leopold, 1978; Harbor, 1994)

In terms of its hydraulics, impervious coverage or impervious area can be distinguished into two categories, effective impervious area (EIA) and non-effective impervious area (non-EIA). EIA comprises those impervious surfaces that are hydraulically connected to the stream – through impermeable storm sewers, culverts, etc. – while non-EIA consists of those impervious areas that drain to pervious ground such as a roof drains onto a lawn or a street draining onto a grass swale, allowing some infiltration of runoff into the soil (Alley & Veenhuis, 1983). The sum of both EIA and non-EIA is the commonly discussed total impervious area or TIA. For assessing watershed impact, EIA is a better indicator than TIA but more difficult to quantify. Different measures taken at the site such as use of swales and ponds would result in different EIA percentages for developments with the same TIA.

Subdivision design affects the amount of impervious surface in many ways, both directly and indirectly. For a start, the density of a subdivision determines the extent of building footprints and driveways, major components of impervious surface, in a subdivision. Impervious surface from the streets serving these buildings meanwhile is influenced by both the design of the street network and the street standards. The design of the street network in turn is very much influenced by the design or layout of the subdivision. Conventional design typically leads to the development of residential subdivisions that completely blanket a parcel with evenly-spaced lots (Figure 2.0). This results from zoning provisions that require minimum lot sizes and widths, and streets that front every lot. In most cases, these streets must have curbs, gutters and storm sewers, increasing the ratio of EIA. In addition, large lots and front yard setbacks necessitate even more pavement to connect garages and front doors to streets. With the entire pavement connected, there is much less opportunity for runoff to soak into the ground. In short, conventional development carries with it a subtle but powerful bias towards maximizing quantity and speed of runoff.

More and more zoning regulations in the USA now have a provision that either mandates or encourages cluster subdivisions (Figure 2.0). Originally, this provision for cluster development is specifically added to reserve a minimum proportion of a site for open space rather than to reduce imperviousness. Of late, however, cluster subdivisions have been promoted as an alternative design to reduce imperviousness (Arnold and Gibbons, 1996; Schueler, 1996; Arendt, 1996). Depending on lot size and the road network, Schueler (1996) claims that cluster development can potentially reduce site imperviousness by as much as 10-50%. In density-neutral provision for cluster development, a developer is still limited to the total units as in conventional development but has the flexibility to place them in a way that is more responsive to the physical characteristics of the site. In spite of more open space, density is maintained by allowing smaller lots than the conventional developments. Smaller lot size that results in narrower and shallower lots also helps reduce the lineal length of streets and total length of driveways. Clustering lots closer together also results in shorter roads. Clustering also enables a better relationship between impervious surface and natural drainageways. Through proper site planning, roads and houses can be located at higher elevation away from natural drainageways, leaving them as open space and giving runoff opportunity to infiltrate while slowly traveling downhill towards the drainage.

Majid (2005) conducted a study on the effects of density and subdivision design on the amount of impervious surface (TIA) and its components, i.e. EIA and non-EIA, for 115 subdivisions in Wake County, North Carolina, USA. The results of the study confirmed findings by other researchers that the amount of imperviousness is positively correlated with density, regardless of subdivision design (Figure 3.0). Even though imperviousness was found to be lower in cluster subdivisions, the amount of imperviousness resulted from streets was not significantly influenced by subdivision design. The level of imperviousness according to density and subdivision design obtained from the study is utilized in this study, with all impervious surface from streets right-of-way assumed to be EIA if the streets are curbed. Detailed description on the methods and results of the impervious study can be found in Majid (2005).

SWAT Hydrological Model
SWAT is a semi-distributed, conceptual, continuous time-step model developed for water resource management and nonpoint source pollution assessment in watersheds and large river basins. A detailed description of the model can be found in Arnold et al. (1998). The model operates on a daily time step and allows a watershed to be subdivided into grid cells or natural subwatersheds. The model process is based on the water balance approach. A distributed SCS (Soil Conservation Service) curve number is generated for the computation of overland flow runoff volume, given by the standard SCS runoff equation (USDA, 1986). Based on surface runoff calculated using the SCS runoff equation, excess surface runoff not lost to other functions makes its way to the channels where it is routed downstream. In urban areas, surface runoff is calculated separately for EIA and non-EIA, avoiding any bias in the calculation of composite curve number. A GIS soil database is used to obtain information on soil type, texture, depth, and hydrologic properties. Since it was developed in the USA, SWAT was designed to take advantage of the readily available STATSGO or SSURGO soil databases. The SSURGO (Soil Survey Geographic Database) is a county-wide large scale (1:24000) soil data set that is more detailed and updated than the large-scale (1:250000) STATSGO (State Soil Geographic Database). Nevertheless, it is still applicable in other countries as long as the soil database is structured to meet its requirements. Accounts of SWAT’s successful applications in other countries are highlighted on its official website (https://www.brc.tamus.edu/swat/).

To take advantage of readily available GIS data for pre-processing of inputs and post-processing of outputs, at least two versions of GIS interface of SWAT have been developed. One is using GRASS, a public domain GIS software, and the other is using ArcView, a widely-used commercial GIS software. The ArcView interface, called AVSWAT, consists of three key components; a preprocessor for generating topographic parameters and model input parameters, an editor for database editing and model execution, and a postprocessor for viewing graphical and tabular outputs. The import/export of data and results between ArcView and SWAT is accomplished through interactive customized menus developed using ArcView’s Avenue routines. The input preprocessor automatically subdivides the watershed (according to topography and user-defined subwatershed threshold size), and then extracts model input data from map layers (soil, land use) and associated relational databases (weather, management) for each subwatershed. The output postprocessor allows the user to display output maps and graph output data in ArcView.

The amount of inputs required for SWAT simulation depends on the purpose of the simulation. However, for basic watershed hydrology simulation concentrating on hydraulics, the minimal input requirements are topography, soil data, land use/land cover (LULC) data, basic climate data (precipitation and temperature) and observed stream flow data for calibration. Meanwhile, important output elements concerning water are runoff, stream flow, evapotranspiration, groundwater, lateral flows, transmission loss, and soil water. Outputs in SWAT are given in three different formats; outputs for the main routing channel (main reach) in each subwatershed, outputs for each subwatershed, and outputs for hydrological response units (HRU) in each subwatershed. A HRU is defined as an area in a subwatershed that is made up of a unique combination of soil and land use types. It is an aspatial entity lumped in its respective subwatershed. Within a subwatershed, a HRU often consists of areas with the same hydrology, soil type and land use characteristics from spatially separated parts of the subwatershed. Since HRU is the primary modeling unit for the model, it assumes that the hydrological routing passages are the same for all the areas that belong to the same HRU even if the areas are distributed at different parts of the subwatershed. In theory, this weakness could partially be overcome by delineating large number of small subwatersheds so that each subwatershed virtually contains only one HRU. In practice, however, this will take a lot of computing resources depending on the size of the watershed and the heterogeneity of land uses, soil types and management practices.

SWAT is currently being utilized in several large projects by government agencies and private consultancies as well as in research at university level. Samples of its applications are summarized on its webpage . SWAT provided the modeling capabilities of the HUMUS (Hydrologic Unit Model of the United States) project to simulate hydrologic budget and sediment movement for approximately 2,100 hydrologic unit areas delineated by the USGS (Arnold et al., 1999). The model was also used by NOAA to estimate nonpoint source loadings into all US coastal areas as part of the National Coastal Pollutant Discharge Inventory (Arnold et al., 1998). SWAT has also been incorporated into BASINS 3.1 by USEPA where BASINS is an interface developed in ArcView to provide state regulatory agencies with the ability to quickly assess water bodies and analyze water quality problems.

STUDY AREA AND OTHER INPUT DATA
The area that has been selected for this study is the watershed of a tributary of Swift Creek, located within the jurisdiction of the City of Raleigh and Wake County, North Carolina, USA (Figure 4.0). The existing land use/land cover of the 4030-acre (1631hectares) watershed is mostly vegetation (forest and agriculture) which accounts for about 63% of the area (Figure 5.0). Residential area, concentrated along the boundary, makes up about 29% of the area, with 90% of the total residential area having a density of 1 d.u./acre or less. Commercial area is located at the northern part of the watershed and constitutes only about 3% of the watershed. The remaining land use is main roads (3%) and water bodies (2%). In term of future development policy, the watershed is part of an area that has been identified by the local authorities as a watershed protection area for maintaining water quality. As a result, the current policy on development within the area limits future development to only residential uses with an average of 1 d.u./acre (City of Raleigh, 2002).

The watershed lies between eighty to one hundred fifty meters above sea level with the high points generally located on the northeast corner and low points on the southeast corner. Figure 6.0 shows the 6m DEM of the area. Based on the SSURGO soil database, there are about 33 types of soil uniformly distributed throughout the watershed (Figure 7.0). Figure 4.0 also shows the stream networks within and around the watershed and the location of the USGS gage that acts as the watershed outlet point. The gage was operated by USGS for only about one and a half year, from May 2002 – September 2003. A climate station recording daily climate data (precipitation, temperature, solar radiation, humidity, etc.) is located at Lake Wheeler Road adjacent to the site (Figure 4.0).

METHODS AND PROCEDURES

Model Set Up
SWAT organizes data into three hierarchic levels: watershed, subwatershed and HRU. Data at the watershed level, i.e. the whole modeling-area level, describe the general configuration of the watershed and the control of water flowing in and out of the watershed. The first to be developed, and most important, data at the watershed level is the watershed boundary delineated using the automatic watershed delineation process embedded in AVSWAT based on the 6m DEM and the 1:14000-scale stream networks. The location of the USGS gage acts as the downstream outlet of the watershed and the resulting watershed is as previously shown in Figure 4.0.

Most physical characteristics of the watershed such as average slopes, routing channel characteristics, point discharges, climate, etc. are dealt with at the subwatershed level. Some of these subwatershed characteristics such as average slopes, for instance, are passed on to the HRUs residing within the respective subwatershed. As such, the smaller the subwatershed the closer its average characteristics are to the actual values, dependent of course on the resolution of the input data such as the elevation DEM and the scale of field variability. There is, however, a trade-off with computing speed as more subwatersheds means longer computing time. For this study, a total of 23 subwatersheds have been created ranging from 22 to 437 acres, giving an average of 175 acres per subwatershed. Variations in topography and land use are considered when deciding the boundary of each subwatershed.

The most refined level of data in term of area is those of the HRUs. Through the GIS overlay process embedded in AVSWAT, HRUs are determined based on user-defined minimum threshold percentage of LULC in a subwatershed and minimum threshold percentage of soil type in the each accounted LULC type. For this study, the threshold values used are 0% and 5%, meaning that all LULC in a subwatershed and all soil types that make up at least 5% of each LULC are included in overlaying the two layers (LULC and soils) to create HRUs. This results in a total of 625 HRUs for the whole watershed with the number of HRUs per subwatershed ranges from twelve to forty eight.

A number of inputs concerning local climate are also necessary in order to run the model. These inputs include daily values of precipitation, minimum and maximum temperatures, relative humidity, total solar radiation, and average wind speeds. However, not all of the data are required in order to run the model. Requirements for climate data depends on the mode of potential evapotranspiration (PET) calculation chosen by the user and SWAT offers four modes of PET calculations. For this study, the Priestly-Taylor method is used and the data required by this method are minimum and maximum temperature, daily solar radiation and relative humidity. All of these data can either be input by the user or generated by SWAT using its built-in weather generator. The weather generator in SWAT generates weather data based on historic data of first-order weather stations in its database. This study, however, uses observed data from the Lake Wheeler Road Field Lab for all of the climate inputs except for a few dates for which solar radiation and relative humidity data are missing. The dates involved are January to May of 1998, the first few days of the 5-year spin-up period. For these dates, the built-in weather generator is used. Since the dates concerned are within the spin-up period instead of the calibration period, the impact of not using the actual field data during these dates on the calibration results is minimal.

Stream Flow Calibration
Once all the required inputs are in place and the model is set up, calibration of the stream flow at the outlet of the watershed is carried out against the daily observed stream flows obtained from the USGS Gage Station Number 0208762750 located at the outlet for the period of one water year from 10/1/2002 to 9/30/2003. The simulation is carried out under the existing LULC condition at the time of the simulation year. The beginning date of simulation is however set at 1/1/1998 since the model requires the initial few years to be used as ‘spinning up’ years. Hence, the results of the simulation from 1/1/1998 up to 9/30/2002 are not used in the analysis and only the results for the period of 10/1/2002 to 9/30/2003 are used.

Calibration of the model is done using the Generalized Likelihood Uncertainty Estimation (GLUE) approach (Beven, 2001; Beven & Binley, 1992). In this study, the hydrological model SWAT is coupled with the GLUE method in LINUX environment using a Python scripting system (Shin, Band & Hwang, 2005). The GLUE method is an extension of the generalized sensitivity analysis of Spear and Hornberger (1980) and is used for calibration and uncertainty estimation of models based on generalized likelihood. The method incorporates information from multiple acceptable model parameter sets within a Bayesian Monte Carlo framework. To do this, a large number of model runs are carried out with each run parameterized with randomly chosen parameter values from uniform distributions across the range of each parameter. The acceptability of each run is assessed in this study by the Nash-Sutcliffe Coefficient of Efficiency (COE) between log values of simulated and observed stream flows. Log values are preferred for their tendency to minimize the influence of stream flow peaks. Nash-Sutcliffe COE is given by:

where Qi is the observed daily flow, Q’i is the modeled daily flow and Q- is the average observed daily flows during the calibration period. The Nash-Sutcliffe COE was first defined by Nash and Sutcliffe (1970) as the coefficient of efficiency that ranges from minus infinity to 1.0, with higher values indicating better agreement between the observed and the simulated flows. Physically, it represents unity minus the ratio of the mean square error of the simulation to the variance in the observed data. Thus, a value of zero for the Nash-Sutcliffe COE indicates that the observed mean is as good a predictor as the model used for simulation, while negative values indicate that the observed mean is a better predictor than the model. A positive value therefore indicates that the model is a better predictor than the observed mean and the model gets better as the value reaches unity. For this study, the parameter set that gives the highest value of Nash-Sutcliffe COE is taken as the optimal parameter set and used in subsequent modeling of the watershed under different LULC scenarios.

Based on past simulation studies using SWAT (Spruill et al., 2000; Eckhardt et al., 2003) and recommendations from the model’s manual (Neitsch et al., 2002), initial batch of parameters that are potentially sensitive and their range of values are identified. These parameters are: i) groundwater delay (days); ii) the minimum threshold depth of water (mm) in the shallow aquifer for return flow to occur; iii) the groundwater “revap” coefficient; and iv) the minimum threshold depth (mm) of water in the shallow aquifer for “revap” or percolation to the deep aquifer to occur. “Revap” is defined as the movement of water from shallow aquifer into the overlying unsaturated zone as a function of water demand for evapotranspiration. With the parameters confined within their pre-selected ranges, a total of 1000 simulations are run, each with random parameter values selected from the ranges. The acceptability of each run is then assessed by comparing simulated to observed stream flows through a chosen likelihood measure, in this case the Nash-Sutcliffe COE. The run that gives the highest likelihood value is then identified. This run is the best simulation to represent the observed stream flow and it gives the optimal parameter set (excluding imperviousness level) to be used in running flow simulations under different LULC scenarios.

Different LULC Scenarios for Flow Simulations and Their Rationales
Once the model has been calibrated, flow simulations are carried out under different LULC scenarios to compare the impacts of different imperviousness levels on watershed hydrology. A set of hypothetical LULC scenarios representing different levels of imperviousness are formulated by assuming additional subdivision development in the watershed. The imperviousness-related variables of the hypothetical additional development and their values as well as the rationales for choosing them are as presented in Table 1.0 below. Incremental effects of imperviousness from these different hypothetical scenarios are compared to the effects of imperviousness from the baseline scenario. The baseline scenario is the existing LULC in the watershed with no additional subdivision development.

Table 1.0: Values of imperviousness-related variables of the hypothetical new development and their rationales

Variable Values Rationales
New Development Area

Density

Subdivision Design

Street Curbing

1. 2400 acres 2. 600 acres

1. 1 d.u./acre 2. 4 d.u./acre

1. Conventional 2. Cluster

1. Non-curbed 2. Curbed

Incremental effects of imperviousness from additional development on stream flows are investigated for two sizes of new development: 1) 2400 acres of new development when assuming all of the undeveloped land is to be developed at the regulated density (1 d.u./acre); 2) 600 acres of new development developed at four times the regulated density, yielding the same number of new units. For baseline scenario, no new development is assumed to take place.

Density of 1 d.u./acre is chosen to investigate the incremental effects of imperviousness resulting from new development developed at the existing zoning density. Density of 4 d.u./acre is chosen to investigate the impact of increased imperviousness from higher density. The density of 4 d.u./acre is arbitrarily chosen from the 1.45-6.0 d.u./acre density range within which subdivision design has been found to have significant impact on imperviousness.

Simulating flows from both conventional and cluster subdivisions will show the impact of different subdivision designs, by way of imperviousness level, on stream flows. At density of 1 d.u./acre, however, subdivision design has not been found to be a factor in determining imperviousness and therefore no differentiation is made between designs in the hypothetical scenarios of this density.

All of curbed streets are assumed to be EIA while non-curbed streets are all non-EIA. The practice of street curbing can therefore directly affect stream flows by increasing the amount of EIA but the degree of the impact is dependent on the EIA amount.

A total of eleven LULC scenarios – one baseline scenario plus ten hypothetical scenarios – are selected on the basis of the variations in the imperviousness-related variables of the new development. Table 2.0 summarizes the characteristics of all the scenarios including the imperviousness levels while Figures 8.0 and 9.0 show maps of some selected scenarios. Important variations among the scenarios are the size and density of new development which determine the amount of TIA, the type of subdivision design which also affects the amount of TIA and the practice of street curbing which influence the amount of EIA.

Table 2.0: Brief characteristics of simulated development scenarios

Indicators of Hydrologic Effects
The hydrologic effects of imperviousness are manifested in the values of the simulated daily stream flows at the downstream outlet of the watershed. Two types of stream flow statistics are used in this study to assess the hydrologic effects, i.e. annual stream flows and daily stream flows. Annual stream flow is the total amount of stream flow in the simulated year and it gives a general assessment of the impact of imperviousness on stream flow. Two components of annual stream flow, baseflow and surface flows, are also evaluated in addition to runoff ratio to further evaluate the hydrologic impact of imperviousness. Baseflow is defined in the model as water for stream flow that comes from ground water and interflow or lateral flow while surface flow is water for stream flow that comes from aboveground runoff. Runoff ratio, meanwhile, is the proportion of precipitation that becomes runoff and its value reflects the intensity of runoff in a given watershed. As imperviousness increases it is expected that the runoff ratio will increase, tipping the balance between baseflow and surface flow of the stream flow in favor of the surface flow. Even though analyzing the impact of imperviousness on long-term flows such as annual flows helps in understanding the balance between baseflow and surface flow and the general impact on stream flows, it tends to obscure the response of each scenario to individual precipitation events involving both minimal and significant rainfalls. The response to each precipitation event, particularly a heavy precipitation event, is important in understanding the influence of imperviousness on peak flows. Thus, the simulated daily stream flows are also analyzed in this study to investigate the daily response of each scenario to precipitations as well as drought. Even though analyses are carried out for each day of the simulation, particular interest is focused on days with high precipitation value. In this study, precipitations of 1 in/day or more are considered high given the watershed’s median and 75th percentile precipitations of 0.3 and 0.73 in/day, respectively. There are a total of 15 days with precipitation of 1 in/day or more with the maximum value of 4.1 in/day.

RESULTS AND DISCUSSION

Calibration and Simulation of Flows
Calibration of the model for daily stream flows between 10/1/2002 to 9/30/2003 indicates sufficient performance of the model which results in Nash-Sutcliffe COE of 0.78. Figure 10.0 shows the time series of the observed and simulated daily stream flows. Considering the facts that calibration is done on daily flows for a period of one year, the agreement between the modeled and the actual flows is satisfactory indeed. Thus the model can be used with confidence in simulating the daily stream flows within that one year period under different LULC scenarios. Simulated daily stream flows of Scenarios 2-11 are shown against the existing stream flow (Scenario 1) in Figure 11.0. The time series of stream flows depicted in the figure seem to fall into three distinct groups. The first is the time series for the existing scenario which have higher volume of low flows and lower peak flows. The second group is made up of flows from Scenarios 2-3 and 8-11 whose time series stay very close to that of the existing scenario with no obvious differences within the group. The third group is formed by flows from Scenarios 4-7 whose low flows are distinctively lower and peaks higher. Differences in flows are noticeable even within the group itself where Scenario 5 which has the highest TIA and EIA generates more extreme flows than the other scenarios. Thus, through observation of the time series alone it is evident that higher imperviousness levels (Scenarios 4-7) generate more impact on stream flows. The following paragraphs will discuss the impact in greater details by looking at other indicators and also at the components of stream flows.


Figure 11.0: Simulated daily stream flows for the demonstration watershed from 10/1/2002 – 9/30/2003 under different LULC scenarios.

Impact on Annual Flows
Table 3.0 below lists the annual values of stream flow, baseflow, surface flow and runoff ratio for each scenario while Figure 12.0 plots these values against TIA values associated with the scenarios. Under the existing land use (Scenario 1) where only 35% of the watershed is developed and 8% of the watershed is impervious surface, the annual stream flow measures 26.4 inch (671 mm) with baseflow makes up 62% of it and the remaining 38% is from surface flow or stormflow. The annual runoff ratio of the watershed under the existing land use observed at its outlet is 14.5%, which means that 10 inches (254 mm) of the total 69 inches (1753 mm) of the precipitation falling within the watershed that year becomes runoff. Even though the annual stream flows differ very slightly among scenarios, there are significant differences in the values of the components of stream flows and those of runoff ratios. As depicted in Figure 12.0, the percentage of surface flow increases with increasing TIA, accompanied by corresponding reduction in the percentage of baseflow. The increasing surface flow results in increased runoff ratio as shown in the figure where runoff ratio increases from 14.5% when TIA is 8% to almost 23% for TIA of 29%.

Table 3.0: Annual values of flow components and runoff ratio under different LULC scenarios

Impact on Daily Flows
In spite of the small differences in annual stream flows between the scenarios, there are significant differences in how some scenarios respond to individual precipitation events. Figure 13.0 shows the plot of stream flows for each individual precipitation event according to scenarios. In general the plot shows that flow volumes increase with precipitation except in a few cases where low stream flows are generated by high precipitations occurring after a period of dry days or high flows by low precipitations immediately following other precipitation events. The data appear to form two distinct groupings of slopes that are significantly different for precipitations equal to or higher than 1 in/day. It seems that precipitations less than 1 in/day are not high enough to make any difference in runoff between scenarios. Figure 14.0 shows the responses of each scenario to precipitations equal to or higher than 1 in/day. It can be seen that Scenarios 4-7, with higher TIAs, show higher stream flow responses to increasing precipitation than the other scenarios. On the other hand, the responses of Scenarios 2, 3, 8-11 to precipitation are not significantly different from the baseline scenario.

For a complete evaluation of the impact of imperviousness levels on daily stream flows, one needs to look at the distribution of daily stream flows during both wet and dry periods. Thus, the distribution of daily stream flows from each scenario is plotted in Figure 15.0 for all days of the simulated year (a), days with measurable rainfalls within the simulated year (b), and dry days within the simulated years (c). For the whole simulated year, there are significant differences in the amount of stream flows among the scenarios. Test of non-parametric Kruskal-Wallis rank test on this data set gives a statistically significant Kruskal-Wallis chi-square of 19.9 (df = 10, p-value = 0.03). The Kruskal-Wallis rank test is employed since the distribution of stream flows is not normal, i.e. positively skewed as can be seen in the plots. Figure 15.0(a) shows the differences in the stream flows where flows from the more impervious Scenarios 4-7 have higher peaks but lower medians. Higher flows in Scenarios 4-7 during precipitation events are confirmed by Figure 15.0(b) which shows only stream flows for days with rainfalls. The difference in the amount of stream flows among the scenarios is tested to be statistically significant (Kruskal-Wallis chi-square = 19.9, df = 10, p-value = 0.03). The differences in amount of stream flows are also tested for days of no rain and are found to be statistically significant (Kruskal-Wallis chi-square = 761.8, df = 10, p-value = 0). The distribution of the flows is as shown in Figure 15.0(c) and it can be seen that during period of low flows (dry days) Scenarios 4-7 record lower flows than the other scenarios. Thus, in comparison to the other scenarios, Scenarios 4-7 are more likely to have lower flows during dry period and higher flows during wet period. This characteristic is common in an urbanized watershed where stormflow periods are brief but with high volumes and rapid recession rates (Konrad and Booth, 2002).

Different levels of imperviousness as represented by the different scenarios simulated seem to have little effect on the annual stream flow of a watershed. For the range of total imperviousness simulated (8-29%), annual stream flows vary only by seven percent. The real impact of imperviousness is actually on the daily hydrographs of the stream flows where the amount of imperviousness determines how a watershed responses to a rainfall event. As imperviousness increases, the amount of surface runoff increases, resulting in higher runoff ratio (Figure 12.0). High surface runoff means less soil infiltration, giving higher stream flow per unit precipitation depth during wet periods (Scenarios 4-7 in Figure 14.0) and lower stream flows during dry periods (Scenarios 4-7 in Figure 15.0(c)). As previously mentioned, this is the characteristic of a flashy hydrograph of an urban watershed where stormflow peaks are high and recession rates are rapid.

The effect of increasing imperviousness on runoff ratio, however, occurs rather gradually in a linear fashion as previously shown in Figure 12.0. Doubling of total imperviousness percentage from eight to sixteen percent, for instance, does not have significant hydrologic impacts on the watershed as shown by Scenarios 2 against Scenario 1 in Figures 11.0 and 14.0. The small difference between the two flows is reflected by the small difference in the runoff ratios between the two scenarios, i.e. about 2.5%. Even if all the roads within the new development were curbed (Scenario 3), raising the amount of EIA from 3.0 to 6.5%, the impacts would remain very much the same as the non-curbed option.


Figure 15.0: Distribution of stream flows according to scenarios for a) the whole simulated year; b) days with precipitation event; and c) days with no precipitation event.

Increasing total imperviousness slightly more than three-fold from eight to twenty nine percent, however, begins to show some marked differences in runoff ratio and flow hydrographs. Under Scenarios 4-7 where total imperviousness is 29% for conventional and 27% for cluster development, runoff ratios increase significantly to above 20% in all four cases and also the balance between surface flow and baseflow now reverses where surface flow is more dominant than baseflow (Table 3.0). The watershed responses to precipitation and the distribution of flows are also significantly different where Scenarios 4-7 tend to have more extreme flows (both low and high) than Scenario 1 (Figure 15.0). A previous finding by Schueler and Claytor (1997) reports that there seems to be a threshold percent imperviousness of around 20% above which the runoff ratio changes rather dramatically. One explanation offered by Brun and Band (2000) is that at low imperviousness the existence of greater vegetation cover moderates runoff production through interception, storage and evapotranspiration. The seemingly linear relationship between total imperviousness and runoff ratio based only on the eleven cases of total imperviousness in this study (Figure 12.0), however, seems to disagree with the finding of Schueler and Claytor (1997). More data points would be useful to refute or accept their finding.

Although significantly different from the outcome of Scenario 1, the outcomes of Scenarios 4-7 do not differ much among themselves. This reveals that density (through comparison of Scenario 1 to Scenarios 4-7) has greater hydrologic impacts than subdivision design (through comparisons of Scenarios 4-7 among themselves). Thus, cluster subdivisions as practiced in the study area have little effects in reducing the hydrologic impacts of imperviousness. The amount of impact reduction, as measured by the reduction in runoff ratio, is minimal. For non-curbed option the reduction is only by 1% from 22% to 21% and the same value for the curbed option, 1% from 23% to 22%.

Another interesting scenario is to investigate what happens to the impact of imperviousness if design capacity is taken into consideration. This is done by comparing Scenarios 8-11 to Scenario 2 & 3. While generating the same amount of new units (2400), Scenarios 8-11 require only one fourth of the new development area required for Scenarios 2 & 3 since they are four times as dense. Interestingly, the results of the simulated flows for Scenarios 8-11 are very close to those of Scenarios 2 & 3 (Figures 13.0 -15.0) but Scenarios 8-11 conserve about 50% of the watershed against 5% by Scenario 2 & 3. The results thus support the position taken by Richards et al. (2003) and Stone Jr. (2004) that design capacity should be taken into consideration when considering imperviousness since considering only coverage percentage alone would unfairly give the wrong impression of high density development.

CONCLUSIONS
All of the flow parameters discussed above – stream flows, baseflow, surface flow and runoff ratio – are affected by the amount of impervious cover which is influenced by density and subdivision design. In addition to its amount, impervious cover also affects runoff through the nature of its connectivity to drainage system, directly or indirectly connected, and through its location within the watershed. The whole concern about location centers on whether or not the distribution of runoff creates enough opportunity for infiltration to occur, given the slope/topography of the area. The amount according to type (connectivity) of impervious surface is handled well in the model since the user has the option of inputting observed values instead of using the default values. Being a semi-distributed model, however, SWAT can only handle the location problem of the impervious surface to a certain degree. If subwatersheds within a watershed are delineated small enough and the land cover resolution is high enough to have only binary impervious and pervious classifications then the model technically could address this location problem. This, however, would require tremendous computing time even for a small watershed like this one, not to mention the time required to prepare the detailed land cover data. Since the LULC information available and inputted into the model is only detailed enough to give the lumped percentage of impervious surface and its type within a subdivision, not the spatial location of the impervious cover, subdivision designs can only capture the difference in impervious surface amount. Having curbed or non-curbed streets meanwhile impacts only the distribution of different types of impervious cover, i.e. EIA and non-EIA. Hence any difference in the results due to subdivision design and type of street curbing can only be attributed to the difference in imperviousness amount and the composition of its types, but not to the location of the impervious surface.

In spite of the above shortcoming, the study successfully shows the impacts of different levels of imperviousness on watershed hydrology and the potential of coupling GIS with numerical modeling in urban planning applications. The impacts are less on the total amount of annual stream flow within a watershed but more on how the watershed responds to precipitation events as reflected in the daily stream flows. For the same precipitation events, a more impervious watershed would have higher peak flows with rapid recession rates during wet periods, resulting in lower flows during dry periods. This is due to higher runoff ratio and less infiltration because of higher imperviousness. The impact of different subdivision designs and the practice of street curbing, as translated into the differences in the amount of TIA and EIA among the simulated scenarios, is rather small. This is because of two factors: 1) the small difference in imperviousness level between the two designs; and 2) the impacts of subdivision design are only represented by the level of imperviousness without taking into account the location of impervious surface. The outcome would likely be different if location were taken into account. Cluster subdivisions, for example, are theoretically more flexible than conventional subdivisions in accommodating site design that would allow for longer period and distance of runoff travel over pervious ground, hence encouraging infiltration.

On the other hand, holding subdivision design and area constant, density seems to have greater influence on flows. This is simply because imperviousness level varies greater across density than between subdivision designs. With or without curbed streets, higher density development generates noticeably higher rate of runoff and lower baseflow than its lower density counterpart. However, if design capacity is taken into consideration, higher density development has almost the same effects on hydrology per dwelling unit as the lower density one. This fact seems to make higher density development more favorable when a situation reveals that developable land is limited and the number of dwelling units is fixed. Thus, comparing imperviousness among residential development options should be done with design capacity taken into consideration to give an unbiased report of the imperviousness level of each option per dwelling unit.

REFERENCES

  • Alley, W. M. and J. E. Veenhuis. 1983. Effective impervious area in urban runoff modeling. Journal of Hydraulic Engineering 109(2):313-19.
  • Arendt, R. G. 1996. Conservation Design for Subdivisions: A practical guide for creating open space networks. Island Press, Washington DC. 184pp.
  • Arnold, C. L. and C. J. Gibbons. 1996. Impervious surface: The emergence of a key urban environmental indicator. Journal of the American Planning Association 62(2): 243-258.
  • Arnold, J. G., R. Srinivasan, R. S. Muttiah and J. R. Williams 1998. Large area hydrologic modeling and assessment part I: Model development. Journal of American Water Resources Association 34(1):73-89.
  • Arnold, J. G., R. Srinivasan, R. S. Muttiah, P.M. Allen and C. Walker. 1999. Continental scale simulation of the hydrologic balance. Journal of American Water Resources Association 35(5):1037-1052.
  • Beven, K. J. 2001. Rainfall-Runoff Modelling: The Primer. John Wiley & Sons, Chichester. 372pp.
  • Beven, K. J. and A. Binley. 1992. The future of distributed models: Model calibration and uncertainty prediction. Hydrological Processes 6:279-298.
  • Brun, S.E. and L. Band. 2000. Simulating runoff behavior in an urbanizing watershed. Computers, Environment and Urban Systems 24:5-22.
  • City of Raleigh. 2002. Raleigh Comprehensive Plan: Swift Creek Watershed Plan. Raleigh, NC.
  • Dunne, T. and A. Leopold. 1978. Water in Environmental Planning. W.H. Freeman. San Francisco, CA. 818pp.
  • Eckhardt, K., L. Breuer and H. Frede. 2003. Parameter uncertainty and the significance of simulated land use change effects. Journal of Hydrology 273:164-176.
  • EPA (U.S. Environmental Protection Agency). 1993.Guidance Specifying Management Measures for Sources of Nonpoint Source Pollution in Coastal Waters. US Environmental Protection Agency # EPA-840-B-92-002. Office of Water, Washington, DC.
  • Harbor, J. M. 1994. Practical Method for Estimating the Impact of Land Use Change on Surface Runoff, Ground Water Recharge and Wetland Hydrology. Journal of the American Planning Association 60(1):95-108.
  • Konrad, C. P. and D. B. Booth. 2002. Hydrologic trends associated with urban development for selected streams in the Puget Sound Basin, Western Washington. Water-Resources Investigations Report 02-4040. U.S. Geological Survey. Tacoma, WA.
  • Leopold, L. B. 1968. Hydrology for Urban Land Use Planning: A Guidebook on the Hydrologic Effects of Urban Land Use. U.S. Geological Survey Circular 554.
  • Majid, M. Rafee. 2006. Estimating residential imperviousness at subdivision level using remote sensing images and GIS. Paper presented at ASIA GIS 6th International Conference. March 9-10, 2006. Universiti Teknologi Malaysia, Malaysia.
  • Majid, M. Rafee. 2005. Residential Imperviousness: Potential Usage of Remote Sensing and Evaluating the Influences of Density and Subdivision Designs. Unpublished PhD Dissertation. University of North Carolina. Chapel Hill, NC.
  • Nash, J.E. and J.V. Sutcliffe. 1970. River flow forecasting through conceptual models, I: A discussion of principles. Journal of Hydrology 10:282-90.
  • Neitsch, S. L., J. G. Arnold, J. R. Kiniry, R. Srinivasan, and J. R. Williams. 2002. Soil and Water Assessment Tool User’s Manual: Version 2000. Blackland Research Center. Temple, TX. 472pp.
  • Rashed, T., J.R. Weeks, D. Roberts, J. Rogan, and R. Powell. 2003. Measuring the physical composition of urban morphology using multiple endmember spectral mixture models. Photogrammetric Engineering & Remote Sensing 69(9):1011-20.
  • Richards, L., G. Andeson and M.K. Santore. 2003. Protecting water resources with higher density developments. Proceedings of the National Conference on Urban Stormwater: Enhancing program at the Local Level. 17-20 February, 2003. Chicago,IL.
  • Schueler, T. R. 1996. Site Planning for Urban Stream Protection. Center for Watershed Protection. Ellicott City, MD. Available online at TOC.htm. Accessed on 10/15/2004.
  • Schueler, T. R. and R. Claytor. 1997. Impervious cover as an urban stream indicator and a watershed management tool. In Effects of watershed development and management in aquatic ecosystems: proceedings of an engineering workshop. L.A. Roesner (ed.). ASCE, New York. pp513-529.
  • Shin, D., L. E. Band and T. Hwang. 2005. Toward more usable environmental model: An experience to integrate RHESSys with CatchLab. Paper presented at The 2005 Annual Meeting of the Association of American Geographers. April 5-9, 2005. Denver, CO.
  • Slonecker, E. T., D. B. Jennings and D. Garofalo. 2001. Remote sensing of impervious surfaces: A review. Remote Sensing Reviews 20:227-255.
  • Spear, R. C. and G. M. Hornberger. 1980. Eutrophication of Peel Inlet II: Identification of critical uncertainties via generalized sensitivity analysis. Water Resources 14:43-49.
  • Spruill, C.A., S. R. Workman and J. L. Taraba. 2000. Simulation of daily and monthly stream discharge from small watersheds using the SWAT model. Transactions of the ASAE 43(6):1431-1439.
  • Stone Jr., B. 2004. Paving over paradise: how land use regulations promote residential imperviousness. Landscape and Urban Planning 69:101-113.
  • Tourbier, J. T. and R. Westmacott. 1981. Water Resources Protection Technology: A Handbook of Measures to Protect Water Resources in Land Development. Washington, DC: The Urban Land Institute.
  • USDA (United States Department of Agriculture). 1986. Urban Hydrology for Small Watersheds. Technical Release No.55. Washington, DC.
  • Ward, D., S.R. Phinn and A. T. Murray. 2000. Monitoring growth in rapidly urbanized areas using remotely sensed data. Professional Geographer 52(3):371-86.
  • Wu, C. and A.T. Murray. 2002. Estimating impervious surface distribution by spectral mixture analysis. Remote Sensing of Environment 84:493-505.
  • Yang, L., C. Huang, C.G. Homer, B.K. Wylie and Michael J. Coan. 2003. An approach for mapping large-area impervious surfaces: synergistic use of Landsat-7 ETM+ and high spatial resolution imagery. Canadian Journal of Remote Sensing 29(2):230-240.