TERM PROJECT
CE394K SURFACE WATER HYDROLOGY
UNIVERSITY OF TEXAS AT AUSTIN

A GIS ASSESSMENT OF NONPOINT SOURCE POLLUTION LOAD IN AUSTIN

By Christine Dartiguenave

Graduate Research Assistant
Center for Research in Water Resources
J.J Pickle Research Campus Bldg.119
Austin, TX 78712
Email : darti@mail.utexas.edu

December 1996

Table of contents

 To delineate the watersheds, the stream network the computation is based on must first be built.

The stream network is based on the flow accumlation grid obtained from the flow direction grid.

Grid: ausfac = flowaccumulation (ausfdr)

The stream network delineation depends on the choice of the threshfold used for flow accumulation.

The streams are delineated by example for a 10,000 cells flow accumulation threshold.
Each cell has an area of 100*100 ft².

Grid: str10000 = con ( ausfac>10000 ,1 )
The grid can then be converted into a coverage.
Grid: stream10000 = streamline (str10000)
Each stream link of the stream network is separated:
Grid: link = streamlink ( str10000, ausdr)
The watersheds corresponding to these streamlinks are delineated using the watershed function.
Grid: shd = watershed (aus_fdr, link)
This grid is then converted into a coverage:
Grid: wshed = gridpoly (shd)

2. From the stations coverages

a. Get the coordinates of the gage stations

There are two types of stations within the area of study.
There are 32 USGS stations within the area of study. Their coordinates can be found in the USGS website USGS Water Data for Texas .

There are 45 EII stations. Their locations have been determined with a GPS system.

b. Create the necessary text files

The coordinates of the stations are given in geographic degrees, minutes and seconds.
To be used to build the coverages, these data have first to be converted into digital degrees :
Decimal Degrees (DD) = Degrees + Minutes/60 + Seconds/3600. The west longitude is negative in decimal degrees.
The coordinates are put in a text file, stat.dat :

1 -98.0806 30.4208
2 -97.9078 30.3917
3 -97.7844 30.3719
4 -97.7861 30.3147
5 -97.9253 30.2961
...
end

c. Generate the point coverages

The point coverages are build in Arc Info from the data in a text file (e.g. stat.dat).
Arc: generate stations
Generate: input stat.dat
Generate: points
Generate: quit
Arc: build stations points
Arc: addxy stations

The result is a point coverage of the stations in geographic coordinates.
It has to be projected to obtain a coverage in state plane projection, which is the projection used for a study at the scale of a city. The input and output parameters of the projection from geographic to state plane are written in a text file called sta_prj :

input
projection geographic
datum NAD27
units dd
parameters
output
projection state plane
zone 5376
datum NAD27
Zunits NO
Units FEET
Spheroid CLARKE1866
Xshift 0.0000000000
Yshift 0.0000000000
Parameters
end

The coverage stations in geographic coordinates can now be converted into the coverage station in state plane coordinates:

Arc: project cover stations station sta_prj #

AML (Arc Macro Langage) is the langage of programmation of Arc Info. Instead of having to type each time the commands step by step, AML enables to write a program which will do it automatically. To build a point coverage, a AML called ptcov.aml is created in a text editor:

generate cover
input [response'name and path of the input file']
points
quit
build cover points
addxy cover
rename cover [response'name and path of the coverage']
&return

The program can be runned in Arc by using the AML command :
&run ptcov

It will prompt for the name and path of the input file and of the point coverage to create.

d. Adjust the coverages

The stations should be located on the delineated creeks (the digitized coverage is not accurate enough). However, it is not always true and the errors have to be corrected. Arcview enables to check the location of the stations but it can not be used to do any modification. Only Arc Info enables to modify the data. The corrections can be made in ArcTools, which can be accessed from Arc or from Grid (in this case, a grid must first be displayed using display 9999). The coverages of the stations must first be converted into a grid (the grid str# of the stream network already exists):

Arc: Grid
Grid: station_gr = pointgrid (station)

Arcview is then used as a interface to display these grids to check if the stations are located on the creeks. If it is not true, then the grid must be modified in order for the stations to be at the "right" place.
Once the grid with the stations is corrected, it is reconverted into a point coverage:

Grid: station_sta = gridpoint (station_gr)

The two following shows the stations coverages built by this procedure:


e. Delineate the watersheds

The point coverages of the stations must first be converted into grids which will be used in the watershed function.

If EII_st is the coverage of the EII station then the procedure is:
Grid: EII_gr = pointgrid ( EII_st )
Grid: EIIwshed_gr = watershed (ausfdr, EII_gr)
Grid: EII_wshed = gridpoly (EIIwshed_gr)

3. Comparison

Here is a comparison between for Waller Creek. On the left are the subwatersheds obtained from stream links for a threshold of 1,000 cells. On the right the green area represents the drainage area for the station at 38th. The purple and the green are the drainage area for the station at 23rd.

A comparison between the USGS data for the drainage area of the 38th and 23rd stations and the values given by the GIS delineation (square miles) shows that the difference is less than 5%.

Stations		38th		23rd

USGS			2.31		4.13
30 m DEM		2.39		4.33		
Error 			3.5%		4.8%

  •     Build a rainfall runoff grid   
    		•

  • 1. Discharge

    Out of the 30 USGS stations in the area of study, 20 have a period of record including at least one complete year. The period of record is based on water-years, which start on October 1 and end on September 30. A water year is named by the year in which it ends. For example, water year 1980 starts in October 1979. The data for annual daily-mean discharge vary a lot from one year to another (e.g. from 1 to 7 cfs for Waller Creek at 23rd).

    Can the prediction model be based on average values?

    If we consider that pollutants concentrations are based only on the land use, then the pollution load for a given land use is directly proportionnal to the pollution load. In this case, using average values is not a good approach.

    For the station at 23rd, the average annual daily-mean discharge is 3.5 cfs. For very wet years (discharge: 7cfs), the pollution loading is greatly under estimated (1/2 of the real load). On the other hand, for dry year (discharge: 1cfs), it is highly over estimated (3.5 times).

    It is then much more realistic to use a probability distribution to predict the discharge.

    The pollutant load is defined as the product of the concentration by the discharge:

    LOAD = Q * C

    Hence
    LN(Q*C) = Ln(Q) + Ln(C)

    The concentrations C are lognormally distributed (they are defined by a median and a coefficient of variation (CV)). Hence LN(C) is normally distributed.
    A propriety of the normal distribution is that the sum of 2 normal distributions is a normal distribution, whose mean is the sum of the means and variance the sum of the variances. If the discharge was lognormal (LN(Q) normal), the natural logarithm of the pollution load would be the sum of 2 normal distributions, LN(Q) and LN(C). Or for a base period of 15 years (1980-1994), lognormal probability distributions fit quite well the observed discharge data which are highly variable (2 orders of magnitude).

    The following equations can then be used to compute the pollution load.

    The parameters defining the concentration and the discharge are the median (50% probability) and the coefficient of variation. Or the equations use the mean and the variance.
    For a lognormal distribution, the mean and the variance can be obtained from the median and the coefficient of variation by using the following relationships:

    The value corresponding to the equivalent normal distribution are found by taking the natural logarithm.

    2. Precipitation

    The other component in the computation of the rainfall runoff coefficient is the rainfall data. The annual average rainfall given by the City to use for loading calculations is 31.08 inches (after EPA procedures and the SYNOP program for hourly data at the Austin airport for the period 1948-1993). Any "event" that did not generate at least 0.05" of rainfall within 6 hours has been deleted.
    The mean annual storm event is one that occurs, on average, 51.8 times a year, has a volume of .60", duration of 7.8 hours, average intensity of 0.106 inches/hour, and with 172.1 hours between storm even midpoints. In Austin, it rains about every 7 days on average.

    The discharge data are highly variable. Or they are directly related to the precipitation. A probability approach has equally to be used to determine the precipitation values which will serve as input to the model.

    Precipitation data can be downloaded from the National Climatic Data Center (NCDC) web site (3 stations for Austin).

    3. Rainfall runoff relationship

    To get the discharge from precipitation values, some rainfall runoff relationships for every watersheds have to be established. These relationships will depend on the land uses (related to impervious cover).

    From the precipitation and discharge data a runoff coefficient, defined as the ratio of the discharge to the precipitation, can be found. For an annual time scale, both contributions of base flow and direct runoff to the discharge are taken into account, so the base flow does not have to be substracted from the gauged streamflow. The following graph represent the variations of the runoff coefficient for the drainage area of the USGS station at 23rd in Waller Creek watershed.

    The runoff coefficient is in a range from 0.21 to 0.46 over a 24 year period with an average value of 0.33. The discharge, and then the runoff coefficient are related to the precipitation and to land use.

  •     Build an EMC grid   
    		•

    The Estimated Mean Concentration values are associated with the land uses. The land use coverage will be given by the City of Austin. It will update the coverage of the Austin Spatial Database CDROM.

    The categories used by the city of Austin are: 

    Current		Impervious
Code	Land Use Categories	Cover Percentage
 100	Single Family		30%-40%
-113	Mobile Home		30%-40%
 200	Multi-Family		45%-80%
 300	Commercial		60%-95%
 400	Office			60%-95%
 500	Industrial		60%-95%
 600	Civic/Educational	30%-70%
 700	Park			5%-15%
 800	Transportation		85%-100%			
-870	Utilities		25%-75%
 900	Vacant/Undeveloped	5%-15%
-940	Water/Lake		100%
    These impervious cover estimates are based on the assumption that most streets and roads are included in the overall land use they support. The streets of a Single Family neighborhood are all included in the Single Family land use along with the homes and yards, and not broken out separately. The Transportation land use is reserved for only the largest of roadways, such as IH-35 and MoPac, or for other uses such as railroads, but not for average city streets. Impervious Cover in Urban vs. Non-Urban Watersheds For each land use category, the impervious cover is presented as a range. In most cases this range is due to the different development densities found in Austin. Single Family developments in the central city tend to be more dense, and have more impervious cover, than similar developments in more suburban areas. It is recommended that in Austin’s more urbanized watersheds, land use categories be assigned impervious cover from the upper end of the range. Conversely, in Austin’s non-urban watersheds values from the lower end of the range should be used. 
    			Urban Impervious	Non-Urban Impervious
Land Use Category	Cover Estimate		Cover Estimate
Single Family			 40%		30%
Mobile Home			 40%			30%
Multi-Family			 80%			45%
Commercial			 95%		 	 0%
Office				 95%		        60%
Industrial			 95%		        60%
Civic/Educational		 40%		        40%
Park				 15%		         5%
Transportation			100%		        85%
Utilities			 50%		        50%
Vacant/Undeveloped		 15%		         5%
Water/Lake			100%	 	       100%

  •    Build an grid giving the pollutant load associated to each cell   
    		•

    The grid is the product of the EMC grid by the rainfall runoff grid.

    Grid: load_grid = EMC_grid*Runoff_gr

  •     NPS pollution loading assessment   
    		•

    NPS pollution assessment for any point is done by using the weighted flow accumulation function.
    Grid: NPS_grid = flowaccumulation (ausfdr, load_grid)

    To observe the results:
    Grid: display 9999
    Grid: mape NPS_grid
    Grid: gridpaint NPS_grid value linear nowrap gray
    Grid: linecolor 4
    Grid: arcs creeks
    Grid: polygons lakes
    Grid: cellvalue *

    The NPS pollution load for any cell is obtained by clicking with the mouse on this particular cell.

    The grid can then be converted into a coverage.
    Grid: NPSint_grid = int (NPS_grid)
    Grid: NPS_cov = gridpoly (NPSint_grid)

    The coverage can be displayed in Arcview by adding the theme NPS_cov. The repartition of the pollution load can be observed by choosing some concentration ranges in the legend editor.

  •     Conclusion   
    		•

    GIS is a very efficient tool for the assessment of nonpoint source pollution load. However, as natural phenomena such as precipitation (rather unpredictable) are involved, and as there are not a lot of data available to build the model, NPS pollution load remains very difficult to evaluate accurately.

    Go back to my Home Page.