Exercise 4: Watershed and Stream Network Delineation

Prepared by David Tarboton, Utah State University.
Purpose
The purpose of this exercise is to learn how to use the TARDEM software for watershed and channel network delineation and analysis.

Contents

Exercise

Start ArcView.  Analysis of grid digital elevation model (DEM) data requires the Spatial Analyst Extension.  In the project window, choose Extension from File menu and check Spatial Analyst from the available extensions.  If you don't have the Spatial Analyst extension you will not be able to complete this exercise.

The data needed is in the file watershed.zip.  Extract all the files to a convenient working directory.  The file reydem.asc contains digital elevation data on a 30 m grid for Reynolds Creek.  It is in an ASCII export format.  In a View window select the File menu Import Data Source.  Select import file type as ASCII raster and click OK.  Select the file name reydem.asc.  Give the output grid the name reydem, selecting the folder where you would like the grid file to be saved.  Respond NO to cell values as integer (This data is actually integer data, but it is generally better to treat elevation data as real).  Respond YES to add grid as theme to the view.  The digital elevation model theme named reydem should now be displayed.

Add a new theme to the view (by clicking on the  add theme button on the top tool bar).  Select the feature data source rcout.shp shown and click on OK to add it to your View. This is the Reynolds Creek Watershed outline.

The data file includes another shape file 17050103.shp.  This is in geographic coordinates (latitude and longitude) and if added to the view will not align with the DEM data displayed which are in UTM coordinates relative to the NAD27 datum (NorthAmerican Datum 1927).  To display and analyze this data together the different coordinate systems need to be reconciled.  ArcView has the capability to project from geographic coordinates to other coordinate systems (but not the other way) so we will project the geographic coordinates onto UTM.

Open a new view, by double clicking on the Views icon  in the project menu. Add the 17050103.shp theme to this view ( add theme button).  From the View menu select properties.  Click on Projection and select category UTM -1927, type Zone 11.  Click OK to the projection dialog and OK to the View Properties dialog.  You should notice a shift in the orientation of the displayed reaches, and that the coordinates in the top left are now UTM (100,000's of meters) rather than latitude and longitude (~100 degrees).

With the 17050103.shp selected from the Theme menu select Convert to Shapefile.  Give a new name (17050103utm.shp).  Say yes to the dialog to save in rojected units.  Say OK to the message about the converted shape not being added to the view.

Close the projected view and return to the first view (with the basin outline, elevation data). Add the new theme 17050103utm.shp to this view.  The streams should match nicely with the watershed outline and topography.

Select the river network within the Reynolds Creek basin outline.  With the17050103utm.shp selected (raised legendbar) choose Select by Theme from the Theme menu.  In the Select by Theme dialog box, select features of active themes that Intersect the selected features of Rcout.shp.  Since Rcout.shp is a polygon shape file this means that all reach segments which are within or intersect the Reynolds Creek Basin will be selected.

Click New Set.  You will see that all of the river reaches that intersect the Rcout.shp are highlighted.  Now with the17050103utm.shp selected (raised legendbar) and Reynolds Creek streams selected (Yellow) choose Convert to Shapefile from the Theme menu.  Provide a name (reystreams.shp) for the new shapefile.  You will see that this comprises only the streams of Reynolds Creek Watershed.  This completes the preliminary data preparation.

To identify and delineate a watershed we will use TARDEM.  (Download and install TARDEM)  Open a MS-DOS prompt (Start/Programs/MS Dos Prompt) and navigate (using cd) to your working directory (e.g. cd c:\giscourse\reydata\).  Now at the prompt enter the command

    tdprepro reydem
This runs a batch file that runs four programs (flood, d8, aread8 and gridnet) that fill pits, compute flow directions and slopes (by the D8 method), compute contributing area and compute grid network order and length attributes.  The files produced are identified by a suffix appended at the end of the name reydem that was given on the command line.  (more details ...)

If you look in the directory where you are working you should see grid files (which are folders in Windows) with name suffixes fel, p, ad8, sd8, plen, tlen, and gord.  These contain pit filled elevations, D8 flow directions, contributing areas, slopes, distance to furthest point upstream, total length of upstream flow paths and grid network order respectively.  (see details...)  Add the grid theme reydemad8. ( add theme button). This is the contributing area grid computed using the D8 method.  Fiddle with the color scheme legend so that you can see the channel network.  (I suggest a legend that has 1-100 transparent, 100-1000 light blue, 1000-5000 dark blue, no data black and everything else deleted).   Drag the legend bars for the stream reach shape file and watershed outline shape file rcout.shp to above the reydemad8 theme so that stream reaches and watershed outline are visible above the contributing area grid theme.  Note the correspondence (or lack of it) between stream reaches and large contributing area values.  It is this correspondence that is the basis for the contributing area threshold method for delineating channel networks.  Note also the black (no data) jagged edges to the contributing area theme.  This is because the contributing area computation checks for edge contamination.  This is the possibility that a contributing area value may be underestimated due to grid cells outside of the domain not being counted.  This occurs when drainage is inwards from the boundaries.  The algorithm recognizes this and reports no data.  This may be overridden by the option -nc in aread8 in cases where you know this is not an issue, if for example the DEM has been clipped along a watershed outline.
Zoom in on the outlet and use Identify  to select a grid cell on the outlet with large contributing area

Write down the outlet coordinates, in this case x=520170 y= 4789800.
Now enter the command

    netsetup reydem -m 1 500 -xy 520170 4789800
This maps channel networks using a constant contributing area with a threshold area of 500 grid cells using outlet coordinates specified.  (The -m 1 designates method 1, which is the constant contributing area threshold. For other methods see...) This results in two new grid files being created, reydemord and reydemsrc.  The first contains channel Strahler stream order, and the second is an intermediate mask file used to indicate the channels mapped.  This command also results in two new text files reydemtree.dat and reydemcoord.dat.  These define the reach linkages and attributed for the mapped channels.  Details of these file formats are given in the TARDEM documentation.  Add the reydemord theme.  Experiment with some other contributing area thresholds.  Report the highest stream order for contributing area thresholds of 100, 500, 1000.

With a channel network defined enter the command.

    subbasinsetup reydem 1
This maps subwatersheds draining to each stream order reach controlled by the order threshold parameter given.  Streams of order lower than the threshold given are stripped off the network before mapping resulting in a coarser watershed delineation.   The output is a grid reydemw, an integer grid giving subwatersheds, reydem.shp a stream network shape file and reydemw.shp a shape file of subwatershed boundaries.  Add these themes into arcview to examine them.  Note the correspondence between the 'value' field in the subwatershed grid, and 'id' and 'wsno' fields in the watershed shape and channel network shape files, to facilitate cross referencing of data.  Note also the channel and subwatershed attributes associated with the shape files.

Now enter the command

    netsetup reydem -m 4 0.4 0.1 0.05 20 -xy 520170 4789800  (If you are using different data the outlet coordinates specified will need to be different)
This delineates a channel network according to the upwards curvature method with threshold 20.  (For a presentation that gives more details on the upwards curvature method see ...)  You may run subbasinsetup to display the subwatersheds from this method if you like.  Now enter the command
    streaman reydem
This produces an output file reydemst.dat that contains statistical properties of the "Strahler Streams".  These are stream segments of the same order.  The file looks like:
Streams with amin =  0.00000E+00
 7
 first link
 last link
 order
 length
 drop
 g. length
 AREA
    530     0     5  0.14018E+04  0.00000E+00  0.12987E+04  0.23912E+09
     74    74     2  0.62699E+03  0.16000E+02  0.57940E+03  0.76140E+06
     58    58     1  0.70456E+03  0.63000E+02  0.56365E+03  0.45360E+06
...
The seven columns are defined in the header as first link number, last link number, order, length, drop, geometric length and area, respectively, for each Strahler stream.  We are going to perform a constant drop analysis, so focus on columns 3 and 5.  Plot the stream drop versus order, for each stream and for the mean of all streams of the same order.  To do this you will need to import this data into software such as Excel, Matlab, Splus or Mathematica (whatever you are comfortable with for plotting).  The result should look something like the figure below.

In this figure the mean drop for each order has been offset from the individual points for display purposes.  The first order stream drops seem to have a mean less than the higher orders (at least visually and discounting the single 5th order stream which is not a representative sample).  This can be tested using the t test for the difference between means.

Here and  are the means of the first and higher order streams respectively.  nx and  ny are the sample sizes and sx and sy the sample standard deviations.  With the above data this evaluates to -5.3.  Roughly speaking the difference is statistically significant when |t| (the absolute value of t) is greater than 2.  Look at a statistics book to be more precise.

Repeat this analysis for channel networks generated with thresholds of 30 and 50, i.e. with the commands

    netsetup reydem -m 4 0.4 0.1 0.05 30 -xy 520170 4789800
    netsetup reydem -m 4 0.4 0.1 0.05 50 -xy 520170 4789800
Find a threshold for which the t statistic measuring the significance of the difference between first order and higher order streams is not significant (i.e. |t| less than 2).  Report this threshold.  (If you are using different data the you may also need to try a few different thresholds beyond the set 20, 30 and 50, say 5, 10, 15, 80, 100, 150.  The right one depends on the drainage density or texture of the topography and resolution of the DEM).  This threshold defines is a channel network that is consistent with Horton's laws.  Other channel networks inconsistent with Horton's laws may be mapping as 'channel' parallel flow down hillslopes. Print a layout of the channel network and watershed mapped in this way.  Determine the drainage density (total channel length/watershed area) and compare the drainage density to the drainage density of the EPA reach files mapped channel network.  Report the drainage density of your mapped channel network and the EPA reach files channel network.

Summary of answers to turn in

Ok, you're done!

These materials may be used for study, research, and education, but please credit the authors and the Utah Water Research Laboratory, Utah State University. All commercial rights reserved. Copyright 2000 Utah State University.