Creating a River Network from the Arcs

in the Digital Chart of the World

by

Kwabena Asante and David Maidment

Center for Research in Water Resources

The University of Texas at Austin

November 1999

Table of Contents

Introduction

The Digital Chart of the World (DCW)  is a compilation GIS data sets covering the whole world. There are 17 data layers in the DCW covering a whole range of fields as diverse as  transportation, hydrology, vegetation and topography. Of particular interest to water resource modelers is data layer number 6 which contains the major drainage features. This data layer consists of two types of coverages; a net coverage, DNNET, containing a combination of line and polygon representation of hydrologic features such as rivers, lakes and swamps, and a point coverage, DNPOINT, demarcating the location of significant hydrologic features such as wells, reservoirs, water falls, rapids, locks, dams and sluice gates. The drainage net coverage, DNNET, is the focus of this particular study. The coverage was developed from 1:1,000,000 scale maps developed for the purpose of inland navigation. The definitions used in a navigational map are rather different from those used in maps produced for hydrologic modeling purposes. Refer to the finding of the DCW & Data Quality project for a more complete description of these differences. A lot of effort must hence be invested in editing the DCW drainage characterization to produce a river representation that can be used for hydrologic modeling. In particular, the streams must be edited to create produce river network connectivity as defined by Jona Finndis Jonsdottir and Kimberley Davis. This involves replacing double lined streams with single lined representations, ensuring that all streams are properly connected to the network and oriented to point downstream, and ensuring that there are no loops in the network. The aim of the study is to define a procedure for creating a single stream network coverage of rivers in the DCW.

Network Connectivity Requirements

The following characteristics were identified by Jona Finndis Jonsdottir and Kimberley Davis  as critical for establishing full network connectivity in a stream coverage:

The Need for a River Network Coverage

The utility of river network coverages such as the RF1 coverage of the United States in hydrologic modeling is well documented. The stream burning procedure required to ensure correct drainage representation in DEM processing requires a river network.  River networks are also used in the preprocessing of many hydrologic modeling systems as seen in CRWR-Prepro and Watershed Modeling System. The most compelling reason for developing global river networks can however be found in the emerging Geo Object Modeling Approach being developed, by ESRI, for the next generation of Arc Info products. These models are an attempt to incorporate additional intelligence into the feature representation process in a GIS modeling environment. It allows more complex relationships to be defined between the different features in a river basin. It also allows behaviours which are characteristic of the different feature classes to be incorporated into the computer representation of the features. In effect, models built using this approach will be better able to mimic the behaviour of the real life hydrologic system. A River and Watershed Geo Object Model is being developed by the Consortium for GIS in Water Resource at the Center for Research and Water Resources. To be able to take advantage of this new Geo Object technology, a river network, with fully network connectivity, is a required input. There is currently no global digitized river coverage which meets network connectivity requirements. The DCW represents the most complete global river coverage. Converting it to meet network connectivity would represent a huge step forward in the effort to build representative hydrologic models globally.

Computer and Data Requirements

    Procedure

    The following is an outline of the procedure devised for extracting the drainage coverages from the DCW CD ROMs and editing them. In this procedure, data is extracted for the whole of the African continent. The Congo basin in Central Africa is subsequently used to illustrate the detailed processing required to achieve full network connectivity in the coverage.

    Selecting the CDROM Tiles

    The ESRI version of the DCW is composed of 4 CD ROM discs covering the areas enclosed by the following lines of longitude:

    CDROM

    EXTENT (longitude)
    DISC1 -90o TO 0o
    DISC2 -180o to 90o
    DISC3 0o to 90o
    DISC4 90o to 180o

    Table 1: Extent of ROM Discs

    The first step is to define the geographical extent of the study area, and to identify which CDROM contains the coverages of that area. The continent of Africa is enclosed by the following lines of longitude and latitude.

    BOUNDARY

    GEOGRAPHIC EXTENT

    Northern Latitude 40o
    Southern Latitude -35o
    Western Longitude -20o
    Eastern Longitude 55o

    Table 2: Geographical Extent of Africa

    Extracting and Merging Coverages

    The CDROMs corresponding to this region are Discs 2 and  3. The data on each of these discs is divided into tiles each covering 5o of latitude and 5o of longitude. A separate directory is created for each tile. The directories are labeled by a location label consisting of two letters and two numerical digits.  Under each tile directory, there are 17 vector coverages representing the layers of information described in the table below.

    Layer Name

    and Number

    Theme

    Abbreviation

    Coverage Name(s)

    1 Political Oceans PO PONET , POPOINT
    2 Populated Places PP PPPOLY , PPPOINT
    3 Railroads RR RRLINE
    4 Roads RD RDLINE
    5 Utilities UT UTLINE
    6 Drainage DN DNNET , DNPOINT
    7 Drainage Supplement DS DSPOINT
    8 Hypsography HY HYNET , HYPOINT
    9 Hypsography Supplement HS HSLINE , HSPOINT
    10 Land Cover LC LCPOLY , LCPOINT
    11 Ocean features OF OFLINE , OFPOINT
    12 Physiography PH PHLINE
    13 Aeronautical AE ASPOINT
    14 Cultural Landmarks CL CLPOLY , CLPOINT , CLINE
    15 Transportation Structure TS TSLINE , TSPOINT
    16 Vegetation VG VGPOLY
    17 Data Quality DQ DQNET

    Table 3: Digital Chart Data Layers

    The Arc Info command MAPJOIN function was used for extracting and merging the drainage coverages of Africa. Click on the following links to view the respective AML programs for extracting the drainage layer data from the respective discs: AML for disc 2 and AML for disc 3. The coverages can also be merged using the Avenue script mergethm.ave. The resulting river coverage is shown in Figure 1 below.


    Figure 1: Merged River Coverage for Africa

    Separating Line and Polygon features

    The merged coverages are initially in geographic coordinates. They must be projected on to a flat surface before any editing can begin. This is because the snap, weed and other feature editing tolerance used in Arc Info are only meaningful if the data is in flat map coordinates. This may not be significant when working with small regions but the effects can be considerable when working with large regions. The projection was done in Arc Info though the same procedure can also be executed in ArcView. The projection parameters for the Lambert Azimuthal Equal Area Projection used in Africa are listed in Table 4 below.

    INPUT
    PROJECTION GEOGRAPHIC
    UNITS DD
    PARAMETERS
    OUTPUT
    PROJECTION LAMBERT_AZIMUTH
    UNITS METERS
    PARAMETERS
    6378137.0
    20 0 0
    5 0 0
    0.0
    0.0
    END

    Table 4: Lambert Azimuthal Projection File for Africa

    The merged coverage was projected using the following Arc Info command:

    Arc: project cover afriver1 afriverprj lm_azafr.prj

    where     afriver1 is the name of the merged coverage

                   afriverprj is the name of the projected output coverage

                   lm_azafr.prj is the projection file

    The projected coverage is form the Congo basin is shown in figure 2 below. Notice the existence of several double lined streams lakes and straight lines marking the edges of map sheets.

    Figure 2: Merged River Coverage of the the Congo Basin

    The river arcs were separated from the polygons in ArcEdit, the Arc Info feature editing module. The results of this separation are shown in figure 3 below.


    Figure 3: Separated River Line and Polygon Coverages of the Congo

    Extracting the Centerlines of Polygon Features

    After separating the river arcs from the polygons, the centerline of double lined streams and lake must now be determined and reincorporated into the river arc coverage. To do this, the polygon coverage is first edited to remove polygons that are formed by the boundaries of non-stream features such as lines of latitude and longitude and coastal shorelines. As shown in Figure 4 below, the resulting coverage contains only the polygons formed by double lined rivers, lakes and reservoirs.

    Figure 4: Polygons of Double Lined Flow Channels in the Congo

     The coverage of double lined channels is then converted from its original vector format to a raster grid. This step allows us to take advantage of the thinning functions available in the Arc Info grid environment. The THIN command allows us to continually chip off the cells at the outer edge of a gridded feature until the feature is reduce to a single chain (or other specified thickness) of cells .  The sequence of commands for this process is as follows;

    Arc: dissolve afrivpoly1 afrivpoly2 myvalue poly

    The field 'myvalue' is a dummy attribute containing a value of 1 for each polygon. The dissolve command removes all the small polygons (islands) contained within larger water bodies. Convert the polygon to a grid with

    Arc: polygrid afrivpoly2 afrivgrid myvalue

    Cell Size (square cell):  100

    Convert the Entire Coverage (Y/N)?: Y 

    Arc: grid

    Grid: setwindow afrivgrid afrivgrid

    Grid: setcell afrivgrid

    Grid: afthin200 = thin ( afrivgrid , positive, nofilter, round, 4000)

    Grid: afriv200 = gridline ( afthin200, positive, nothin, nofilter, round, value)

    Grid: quit

    Figure 5: Raster and Vector Derivatives of Thinning

    The Figure 5 above shows a section of the double lined river after conversion into raster format (in green) and the result of the thinning process in blue. The thinned raster representation can then be converted back into a single line river representation as shown in part b of the figure. From the image above, it is clear to see that centerline thus created has a lot of sharp bends which would result in a longer than expected river reach length if used in hydrologic modeling in its present state. The GENERALIZE command is consequently used to remove any  unnecessary nodes. 

    Figure 6: Effect of generalization on Vectorized Centerlines

    The Figure 6 above shows the effect of generalization. The original centerline shown in blue is generalized to a more representative straight line representation. Note that the generalization is sensitive to the weed tolerance defined. For the centerlines on the water bodies in Africa, a threshold of 200 meters gave the best result. The syntax of the generalize command is as follows;

    Arc: generalize afriv200 afcenterln 200 pointremove

    Establishing Network Connectivity

    The centerlines can now be reincorporated into the river network by simply selecting all the arcs in the stream centerline coverage and putting them into the river line coverage.  The resulting coverage must then be edited manually to ensure all arcs are appropriately connected to the network. Refer to ArcView based scripts developed for the CRWR River Network Preprocessor for trouble shooting and editing tools. The Arc Edit commands listed below may also be used. 

    Initiate the Arc Edit environment with the command

    Arc: arcedit

    Arcedit: display 9999

    Arcedit: edit rivercov arc

    specifies the edit coverage and feature type. The edit distance, a measure of how close one has to get to a feature before being able to select it, is specified with

    Arcedit: editdistance

    Similarly, the snap distance is measure of how close two features must be to each other before they are automatically snapped together. The snap distance is specified for nodes and arcs respectively with

    Arcedit: nodesnap closest *

    Arcedit: arcsnap closest *

    Some commonly encountered mistakes are listed below along with ways to correct them.

    1) Undershoots occur when an arc which is supposed to intersect another arc stops short of its target. As illustrated below, the way to correct an undershoot is to select the arc and extend it using the commands listed below.

     Arcedit:   select   extend

    2) Overshoots arise when an arc extends past the arc it is supposed to be linked to. Overshoots can be corrected by coverting the arc to an undershoot and extending as described in (1) above.

     Arcedit:   select    split   select   delete   select   extend

    3) Missing arcs can be replaced by on screen digitizing using the ADD command.

     Arcedit: 

     add

      4) Mismatched nodes can be brought together by setting the edit feature to node and moving one of the nodes to coincide with the other.

     Arcedit:   ef node

    move

     To save edits (do this often to avoid losing edits),

    Arcedit: save

    Build the resulting line coverage to reestablish arc topology and to update the values in the arc attribute table.

    Arc: build rivercov

    It is important to be able to check make sure that the network is fully interconnected. This can be done by running a trace in ArcPlot as follows:

     Arc: arcplot

     Arcplot: display 9999

    Set the map extent to that of the river coverage with
     Arcplot: mapextent rivercov

    To display the arcs in the river coverage, type
     Arcplot: arcs rivercov

    Begin the trace with
     Arcplot: trace direction rivercov upstream downstream *
    (select an outlet from the display by clicking on it)

    To display the results of the trace, type the following sequence of commands
     Arcplot: linesymbol 2
     Arcplot: readselect downstream
     Arcplot: readselect upstream keep
     Arcplot: arcs rivercov

    The arcs that are not connected to or pointing away from the outlet will be shown in white. All other arcs will be shown in red. Figure 7 below showns and example of misoriented arcs in a network.

    wpe1.jpg (30096 bytes)

    Figure 7: Congo Basin Network with Misoriented Arcs in White

    To correct the orientation of misoriented arcs, open the coverage in Arc Edit as before.

    Arc: arcedit

    Arcedit: display 9999

    Arcedit: edit rivercov arc

    Select all the misaligned arcs using the select many command

    Arcedit: select many

    Now change the orientation of all selected arcs with

    Arcedit: flip

    The orientation of the arcs will now be corrected so that they point towards the outlet. Save the coverage and exit Arc Edit as before. Redo the trace to confirm that all arcs are now connected to the outlet.

    The editing of the editing of the DCW stream will require a major effort involving several individuals. However, the results of applying a similar approach to clean up the ArcWorld dataset in the Congo Basin are shown in Figure 7 below. The resolution of the river representation in the ArcWorld dataset is 1:3,000,000.  


    Figure 8: Congo River Network from Arc World

    The underlying gray line is the original ArcWorld data set while the edited river network of the Congo is shown in blue.

    References