Toar T. Schell, Miguel F. Acevedo, Fred C. Bogs, James Newell, Kenneth L. Dickson, Institute of Applied Sciences, University of North Texas, Denton TX., and Foster L. Mayer Environmental Protection Agency, Gulf Breeze Research Laboratory, Gulf Breeze FL.

Assessing Pollutant Loading to Bayou Chico, Florida by Integrating an Urban Stormwater Runoff and Fate Model with GIS

ABSTRACT

This paper discusses the integration of Geographic Information Systems(GIS) and EPA's Storm Water Management Model (SWMM) as part of a watershed approach to assessing the ecological health of Bayou Chico, which is a sub-estuary of Pensacola Bay. Bayou Chico is the receiving water body of small, mostly urbanized watershed located in southern Escambia County, Florida. Both natural channel flow of precipitation and dry weather flow through a managed drainage system contribute to pollutant loading to the bayou. The transport and fate of pollutants are thus represented as water flow through these systems. A GIS database and the use of remotely sensed satellite images are combined to determine surface characteristics, storm drainage systems, area and slope of the watershed. These data layers are in turn linked to SWMM which mathematically represents these physical characteristics and uses this information to determine both runoff and pollutant loading. We discuss data required for SWMM, and its acquisition and use in the GIS. We will emphasize the GIS-model linkage with special attention given to model calibration techniques.

INTRODUCTION

In addition to the natural channelization of a given watershed, human endeavors have made the transport of water over and through a given watershed all the more impactive. Increased surface imperviousness concurrent with urbanization often result in the need to re-route stormwater flow. The introduction of stormwater drainage systems can expand the size of drainage area to a particular receiving water body. This results in the potential for increased pollutant loading to endpoints of such systems as pollutants are transported as a function of runoff. Physically based, spatially distributed models can be utilized to as a means of simulating dynamic runoff response for urban catchments. This is especially useful in cases where continuous measurements of discharge data are lacking , increasing calibration difficulties. Both structural and non-structural urban stormwater management strategies may be evaluated with greater confidence using physically based models, requiring a greater emphasis on database information (Meyer, 1993). The complexity of database information representing attributes for a given location necessitates the use of a GIS. Georeferenced features in a GIS are defined by points, lines or polygons, and may have several attributes attached to them each representing a different component of the hydrologic processes. The result of this is that model parameters can also be considered attributes for a particular map feature (Vieux and Needham, 1993). And that the integration of physically based mathematical models and GIS has become a common method for use in ordering the complexity of data necessary for the understanding of such systems (Moore et al., 1993).

Recently, efforts have been made by the University of North Texas (UNT) and the Environmental Protection Agency, to assess the ecological health of Bayou Chico: a small sub-estuary of Pensacola Bay, Florida. As part of a watershed approach to biological/chemical analysis of the bayou, a modeling component focuses on predicting the fate and transport of non-point-source pollutants in stormwater runoff entering the bayou. This is done through the integration of a physically based hydrologic model: the Environmental protection Agency (EPA) supported Storm Water Management Model (SWMM 4) (Huber and Dickinson, 1988), with a GIS: Arc/INFO 7.0.2 (ESRI, 1994). This paper describes criteria for model selection with a brief discussion of current SWMM/GIS interfaces available, as well as model calibration/verification techniques. We will also discuss a broad overview of SWMM integration with Arc/INFO based GIS analysis.

STUDY AREA

Located next to the city of Pensacola, Bayou Chico is the receiving water body of a small, mostly residential and urban watershed located in southern Escambia County, Florida. The watershed of Bayou Chico is defined primarily by the natural surface flow feeding its shoreline and contributing streams. Additional watershed area is included by the channelization of water through stormwater drainage elements that may not correspond to overland surface flow. We model watersheds Sub-bas 2 (109 ha.) for model calibration and SB-4 (556 ha.) for model verification. Both sub-watersheds comprise of primarily residential and commercial land use. Both drain into fresh water streams at the northeast and northwest portions of the bayou respectively. A large portion of the drainage area consists of a wetland/swamp which functions as a water purification system for the bayou (Pratt et al., 1993).

SWMM SELECTION

Model selection for the Bayou Chico study was based on the following criteria: (1) Must be physically based to take advantage of information derived from GIS analysis; (2) Model must easy to use, capable of simple to complex modeling with a minimum of program development; (3) Well documented with a history of use; (4) Model to be capable to model urban hydrologic processes the output of which can be exported estuarine fate and transport model; (5)must be cost effective.

First developed in 1971 under the supervision of the EPA, SWMM has a long track record that is well documented. Both the fortran code and executables are available at no cost, as is some documentation on file input formats for the various model modules or "blocks". Documentation while disparate, enables the user accessibility to the model's fundamental capabilities. A more detailed Users manual is available at a relatively small cost (see appendix A).

SWMM is a mathematical abstraction of the physical characteristics associated with an urban watershed. Topological characteristics including surface characteristics, stormwater drainage structures and gutters are described by the model. As such it is well suited to parameter input resulting from GIS analysis. The SWMM model is capable of a range of complexity in modeling from runoff in a single watershed with no pipe or channelization network, to that of a more complex system of watersheds and sub-watersheds, each feeding a pipe network with storage and treatment facilities. SWMM core programs are the modules or blocks: RUNOFF, TRANSPORT and EXTRAN. RUNOFF and TRANSPORT are capable or routing surface/groundwater flow and pollutant transport. Moreover, the transport block is capable of modeling dry weather flow and infiltration into sewer systems. The EXTRAN block while not capable of pollutant transport modeling is capable of complex hydraulic routing (Huber and Dickinson, 1988).

SWMM/GIS INTERFACE CONSIDERATIONS

SWMM has historically not had a user friendly interface. The integration of SWMM and Arc/INFO on a UNIX based platform has been recently been developed. SWMMDuet (Curtis, 1994) uses Arc Macro Language (AML) point and click programming as an interactive process for database management, GIS analysis and subsequent calls to run the SWMM program. The demonstration files which come with the program are designed to show how the database items for each geographic layer are to be structured. Pre-existing Arc/INFO coverages must be converted to this format with specific coverage and item names. The program when implemented should provide a seamless method of integrating the GIS with the model. Arc/INFO AML point and click widget programming tends to run slow on the Sun SPARC-2,5 and 20 workstations on which we tested the program . According to the author, the program runs under Arc/INFO 6.0 and a system with a IBM or DEC mainframe; however, there was some difficulty in getting SWMMDuet to run successfully under Arc/INFO 7.0 and a Sun Operating System. (Curtis, 1994). We were unable to successfully compile the Fortran77 code that came with SWMMDuet on our Sun SPARC-2 and 20 workstations and there was insufficient time on our part to debug the code to fit our system. For this reason, SWMMDuet was not used in our study, even though it would have been ideal.

The EPA has developed a windows based interface which allows for manual input of parameter values in a series of spreadsheet-like pages. Hydrographs and pollutographs for up to three outfalls along with measured data, as well as a plot of measured vs predicted data for a single outfall may be viewed after each run. New users and those wishing to use it as a screening tool will find the program invaluable. It also enables veteran users to systematically build the model with increasing complexity with model input displayed in a clear fashion. From within the Windows interface, there is no way to run in a batch mode, therefore changes in input must be carried out interactively. Users familiar with SWMM text format may find this method more cumbersome than SWMM run in batch mode both for sensitivity or calibration purposes. For screening purposes or for use as an initial data entry interface it is extremely useful. The DOS executable that is run from Windows is capable of batch-mode programming outside the Windows interface. An unofficial DOS based SWMM version is also available from the same source as SWMMDuet. It is recompiled to take advantage of extended memory and is constantly being improved (see appendix A).

MODEL/GIS INTEGRATION

Previous Studies

Our study builds on a previous study conducted by the Northwest Florida Water Management District (NWFWMD). This study used SWMM to calculate the mean flow volume and mean peak flow. These were then used to calculate pollutant load based on land use class. The use of GIS in this study was limited to land use analysis determining percent impervious surface (Pratt, 1989). Our approach seeks to use GIS to parameterize the model as much as possible, including watershed boundaries as well as stormwater routing and pollutant loading. The use of GIS analysis with the delineation of watersheds and the routing of storm runoff has been previously attempted by Stuebe and Johnston, (1991). Recent efforts have expanded the GIS/model integration to include water quality evaluation for urban watersheds. (Tim, 1992, Shea et al., 1993, Moore et al., 1993). Ross and Tara (1993) have identified five steps for model processing with GIS: (1) digital data gathering; (2) GIS operations; (3) model input data processing; (4) hydrological Simulation (5) output processing.

METHODS

Digital Data Gathering

Digital databases and their sources have been well documented (Moore et al., 1993). Such data range from freely available via anonymous file transfer (FTP) over the Internet, to manual digitizing data from hardcover maps ( fig,.1.a,b,d).

The SWMM RUNOFF block can be broken into a hydrological component (Table 1.1), and a pollutant loading component. Both have parameters that may be retrieved from spatially based data. Pollutant loading is partially determined either by land use class or gutter length. Gutter length can be derived from digitized road layers or are available for some metropolitan areas in the U.S. Census Tiger Data files. Tiger roads are digitized at a scale of 1:100,000 and as such may have too many inaccuracies for more detailed modeling efforts. Maps from a local entity may serve this purpose better; however this data is available at little to no cost, and is already in a georeferenced digital format.

Image processing software such as ERDAS/IMAGINE 8.2 (1994) allow for the georeferencing of images interactively. RMS error is reduced by selecting ground control points (GCP) and moving them on screen (or dropping them) until an acceptable reduction of error is reached. This software was used to georeference Landsat TM (Fig 2.a) and SPOT images from August 1989 as well as perform a principle components analysis on the TM image using all of its spectral bands. The resulting image was then classified into land use categories and stored as an image file which was exported to Arc/INFO (Fig. 2.b).

We have used scanned 1:24000 scale USGS 7.5 topo maps in TIFF format and imported this to Erdas/Imagine and georeferenced this to the hardcopy version on a digitizing tablet (fig. 1.a). Using ARCEDIT module of Arc/INFO and the georeferenced topo image as the backdrop, data layers derived from the topo image were digitized on screen ( fig. 3). Advantages of this method are: (1) Speed in processing - map is placed on digitizing tablet only once, reducing the possibility of map movement; (2) more control in reducing error; (3) zoom in/out function allows for more accurate digitizing, e.g. closely spaced elevation contour lines; (4) digitizing errors are more apparent; thus, are easily edited.

GIS Operations

GIS operations consist of Data manipulation to derive parameter values from various georeferenced data layers. Coverages are either digitized or derived from the combination of several coverages. The Arc/INFO TIN modeling module is used to create a Triangulated Irregular Network (TIN) from DEM data or digitized points along elevation contours found on 7.5 minute Topographic Maps (Fig 1-C). When overlaid by channel networks or topology the TIN will serve to determine overland flow in the form a slope or aspect towards particular inlets or catchbasins as well as the slope of sub-watersheds and channel networks. Care should be taken when using TIN modeling in areas where the topology is relatively flat. Vieux and Needham (1993) have noted ambiguous results under flat conditions when parameters such as aspect and receiving cell location are being predicted. When encountering ambiguity, they opted for determining such parameters manually.

As a vector based GIS, Arc/INFO has tools that allow for querying of polygons for areal extent, perimeter, length, width and other associated attributes. AML programming is used to query named Arc/INFO coverages for SWMM parameter data and to write the information to an ASCII text in the SWMM input file format. GIS AML programming as planned is to be done on a layer by layer basis increasing with model complexity. To date only preliminary AML programs have been completed for a single watershed (sub-bas 2) with a single pipe network.

Model Data Input Processing

Model Input data consist of values derived from GIS analysis and precipitation data. Long-term simulations may require National Weather Service hourly or 15min rainfall data. A single storm event may also be modeled, and in such cases data from raingages may be entered manually or imported from an ASCII file. Output from the GIS analysis is in the form of a SWMM input template. Parameter values used for model calibration (e.g. impervious area Mannings friction coefficient) are written as a minimum value within the range for that surface type. This value then can be incremented in batch-mode and subsequently tested for contribution to model error. The Awk script parsing language program (appendix A) is used to create duplicates of this file each with a single parameter value changed by the incrementation value chosen for the particular parameter. The results of the Awk program are a user defined number of SWMM input files (runs) along with a batch file that calls these files to the SWMM executable (fig. 4).

Hydrologic Simulation

The SWMM program when implemented in batch mode is designed to produce a series of runs with several parameters used to calibrate flow to a particular endpoint. A secondary Awk program is designed to take SWMM output files and parse out flow and pollution data. It also queries the output file for a lower limit for flow and boundary times for peak flow. We use Total runoff volume and time to peak flow as our benchmarks. Files which meet the criteria are passed along with measured data to a post-processing program for statistical comparisons with including regression and correlation analysis. Original files are archived in case they need further examination. This has proven to be an effective method for a quick sensitivity analysis for the calibrating parameters (fig. 4).

Preliminary runs have been performed on a simplified watershed with a single pipe network. Parameters from table 1-1 that have been used to calibrate the model have been the watershed width, pervious and impervious manning's coefficients, and three Green-Ampt infiltration parameters.

Output Processing.

The present graphic output from SWMM output files are designed for a line plotter and as such does not take advantage of modern printers. Traditionally one has had to import hydrograph and pollutograph data into a spreadsheet or other graphical program. Five goals were established for graphical outputs from SWMM: (1) View hydrographs, pollutographs and statistical graphics in an automated fashion; (2) View the results of multiple SWMM runs sequentially; (3) Produce graphs of presentation quality; (4) do this with minimal programing; (5) be cost effective

We take advantage of a publicly available graphics plotting program GNUPLOT, which has been developed for DOS, WINDOWS 3.x and UNIX (see appendix A for sources). We used a C program to read multiple combined data files, calculates regression and correlation statistics and also write a GNUPLOT script file to produce graphs for each run designated. Output files can be printed or viewed sequentially (fig. 5). This is an extremely effective method, both in terms of time and cost for viewing and publishing SWMM output.

Conclusions

Preliminary results have shown that a SWMM GIS linkage for all spatially oriented attribute data is feasible. Results of the initial runs can be seen in fig. 5, notable are the time to peak and overall shape of the curve. Predicted total runoff volume (area under the curve) however, is much less than observed. Work still to be completed include the subdivision of the Sub-Watershed into drainage areas based on more than one inlet location. In this regard, results from initial TIN modeling have been ambiguous, primarily due to the flatness of the upper half of the Sub-basin. Pollutant loading has not been calibrated. Validation for the model will be performed on sub-basin 4 in the near future. To date completed tasks include the collection of all digital data, rudimentary AML programming to describe watershed parameters, and an automated screening and graphic interface with SWMM output.

APPENDIX A:

Internet Resources - no-cost programs

SWMM

The model is available in various formats. A windows 3.x version is available for remote download from the EPA World Wide Web site:

http://earth1.epa.gov/SWMM_WINDOWS/.

A newly compiled unofficial DOS version (4.31) of SWMM available from Oregon State University:

ftp.engr.orst.edu/pub/swmm/pc

This takes advantage of extended memory not available the official EPA version. (It doesrequire a math co-processor). This version is not supported by the EPA.

The unix based Arc/INFO-SWMM program

SWMMDuet

Also available at ftp.engr.orst.edu/pub/swmm/workstation along with a users Manual.

AWK

An awk version (awk, nawk, gawk) is found on most unix operating systems (including LINUX) A Dos version of Awk is available via anonymous ftp from:

http://www.acs.oakland.edu/oak/SimTel/msdos/awk.html

GNUPLOT

GNUPLOT is available from:

http://www.acs.oakland.edu/oak/SimTel/msdos/plot.html for both the dos andwindows versions.

A manuual is also available.

Dos: gpt35doc.zip (Image) 93/10/15, 600071 bytes

PostScript documentation for gnuplot 3.5 gpt35exe.zip (Image) 93/10/15, 552716 bytes

gnuplot 3.5: 2D/3D plots of data & fcns gpt35src.zip (Image) 93/10/15, 740888 bytes

MS Windows GUI version of gnuplot 3.5 Complete source of gnuplot 3.5 Windows: gpt35win.zip (Image) 93/10/15, 419933 bytes

References

Curtis, T.G. (1995). SWMMDuet Users Guide. Unpublished. Available from Oregon State University via anonymous ftp.engr.orst.edu/pub/swmm/workstation 42pp.

Huber, W. C. and Dickinson, Robert E., (1988), Storm Water Management Model, Version 4: User's Manual. Environmental Research Laboratory, Office of Research and Development, U.S. Environmental Protection, Athens, Georgia. 569 pp.

Meyer, S.P., et. al. (1993). "Geographic Information Systems in Urban Storm Water Management." Journal of Water Resources Planning and Management, 119(2), 206-228.

Moore, I.D. et. al. (1993) "GIS and Land-Surface-Subsurface Process Modeling." in Goodchild, M.F., Parks, B.O. and Steyaert, L.T. (eds.) Environmental Modeling and GIS, Oxford University Press, New York, pp. 196-230.

Pratt, T.R., et al. (1993). Stormwater Assessment of the Bayou Chico Watershed, Escambia County Florida. Surface Water Improvement and Management Plan: A Comprehensive Plan for the Restoration and Preservation of the Pensacola Bay System. Northwest Florida Water Management District (NWFWMD), Water Resources Special Report 93-7, Havana, Florida.

Ross, M.A., and Tara, P.D. (1993). "Integrated Hydrologic Modeling with Geographic Information Systems." Journal of Water Resources Planning and Management, 119(2), 129-140.

Stuebe, M., and Johnston, D. (1990). "Runoff Volume Estimation Using GIS Techniques." Water Resources Bulletin, 26(4), 111-116.

Tim, U.S., Mostaghimi, S. and Shanholz, V.O. (1992). "Identification of Critical Nonpoint Pollution Source Areas Using Geographic Information Systems and Water Quality Modeling." Water Bulletin, 28(5), 877-888.

Vieux, B.E., and Needham, S. (1993). "Nonpoint-Pollution Model Sensitivity to Grid-Cell Size." Journal of Water Resources Planning and Management, 119(2), 141-157.

Toar Schell,

Institute of Applied Sciences

University of North Texas

Denton Texas.

Email: schell@unt.edu

Phone (817) 565-4350