Streams

Given a coverage of streams and and a coverage of contour lines (with an optional spot elevation coverage), we want:

• a topological tree of streams, all pointing downhill
• shreve and strahler order, and order from outlet.
• an inventory of outlets and sources
• elevations of stream junctions, based on a linear interpolation along the main stream segment between contour lines.
• length and slope of streams based on contour lines and junction elevations.
• the best possible DEM, guided by the stream network.
• a grid of contributing area, tied within one pixel to the stream network.
• the ability to interrogate the stream map for contributing area at any point

• new 1/2006 proffix2.aml Improved version of proffix.aml. I am still not happy with this version. The algorithm used for tracing through sinks was designed for small sinks, and it behaves badly in long sinks such as I find in SRTM DEMs. It can make bad decisions, and it sometimes enters infinite loops. (This macro works well with the "hydrosheds" SRTM DEMs, which have no sinks.) I had made the aml code to complicated for me to maintain, so I started to rewrite it in python. However, ESRI does not provide an equivalent to [show cellvalue], so I cannot update the macro.
• new 7/99gridnice.aml a macro to grid lines better by gridding at a high resolution and resampling.
• new 6/99 proffix.aml a macro to trace longitudinal profiles on DEMs, and to repair profiles and DEMs
• [missing]We have no tool (except arcedit) for drawing centerlines down lakes and two-bank streams. (and deleting psuedo-nodes)
• flipperq.aml: This clips the rivers to their outlets, sets the rivers pointing to the outlets, calculates order from source, and reveals topological problems. It also inventories outlets and sources. This version uses a cover named mouthline to clip and define the outlets. It is most appropriate for quads with multiple basins.
• flipperp.aml: This differs in function from flipperq.aml in that the mouth[s] is selected interactively. It has the bonus feature of calculating total length to the mouth.
• jorder.aml: If you already happen rivers pointing downstream, this will number from the mouth and total areas.
• shreve.aml: Calculate shreve and strahler order.
• nodatts.aml: a little utility to transfer arc attributes to nodes.
• topgrida.aml: calculates elevations of all nodes and slopes of all streams.
• streamslope.aml: This does the same thing as topgrida.aml, but it does not use a DEM at all. beta version
• streamslope2.aml: Version of streamslope that calls a java program.
• Sloper.java: Preliminary version of program that swallows an aat file and calculates junction elevations, and slope.
• Aslop.java: Much improved version of the above program. The attribute list is still hardwired, but it implements a Arc class, and has flexible file size. updated 11/21/97)
  See my java page for more goodies.
• Aslop2.java: Version of Aslop that calls a flexible arc class that can have any attributes.
• newc.aml: builds buffers around all streams and intersects them with the contours. Discards portions of the contours between buffers and portions of the buffers with the same to and from elevations. Decrements the elevations of the stream nodes, and uses them with contours, streams, and buffers to build a DEM.
• step4.aml: grid the rivers, calculate the distance from the mouth along the grid, and use it as a surrogate for elevation. Subtract a large elevation and combine with the real elevation grid. The resulting flowdirection grid matches the river. Thinning the gridded river is done in stages. First we grid at a finer resolution and count the number of small river cells within a large cell. Then we thin by pruning everything which is not part of a route from a source node.
• rivgrid.aml: New, improved method of rasterizing rivers.
• something: do calculated non-mapped tribs. Intersect with thinned version of reconstituted rivers. destroy dangles. extend downstream. never mind, nobody needs this at this time.

Let me start over and describe general paths:

Start with mapped streams and contours (and maybe spot elevations):

• Initial QA check on contours. Run aml or java program to check for connected contours with different elevations.
• (if necessary) Make sure streams are a topological tree (no cycles).
• (optional) Edit streams if they are grossly out of register to contours.
• Intersect streams with contours. Flag all zero-slope and neg-slope segments.
• (if necessary) Manual edit. Flat segments are either contour labeling errors or streams crossing contours. Redo the the stream intersections.
• Identify stream source nodes for later use.
• Rasterize the river. Use fancy techniques to get thin segments.
• Do costallocation (or fill) and calculate distance to mouth.
• Create a river coverage of straight and diagonal segments. Use the streamline algorithm, considering distance-to-mouth.
• Also create a point for each river cell.
• Trace from source nodes to mouth(s), thereby pruning fluff on the river coverage. The resulting river can (if desired) be gridded and vectorized without undergoing changes.
• (optional) Create stream buffers to influence the creation of a DEM.
• Create a DEM from contours, (real?) streams, stream junctions, and (if any) spot elevations.
• Make a bogus DEM by merging the river and real DEM. Calculate flow direction. Discard the bogus DEM.
• Calculate flowaccumulation.
• Mask to the river grid. Write out all flowaccumulation points.
• For each orthoriver segment, look at all touching flowaccumulation points. Discard the point that matches the tonode, and take the maximum of the rest.
• Transfer values from the orthorivers to the natural rivers.
• If any segment falls through the cracks, sum the upstream values. If there is no upstream arc, look downstream.

  Start with contours (and maybe spot elevations)

• Initial QA check on contours. Run aml to check for connected contours with different elevations.
• Build DEM from contours (and maybe spot elevations) The maximum cell size for 40-foot contours in the mountains is 10 meters.
• Fill sinks. (Examining the difference between the original and the filled DEM can help identify contour coding errors.)
• Calculate contributing area. (Use C-program for distributed flow.)
• Calculate [ARC/INFO] 9-cell slope and select channelhead threshhold.
• Define channelheads based on area × slope˛. (or use another power of slope) A filter can eliminate adjacent channel heads. Use C program to complete discontinuous channels. A byproduct (not yet implemented) is a file of channelhead locations.
• (optional) Edit streams if they don't look right. There are artifacts of topogrid's mediocre algorithm, amplified by sink filling, as well as problems resulting from cell size.
• Rasterize the river. Use fancy techniques to get thin segments.
• Do costallocation (or fill) and calculate distance to mouth.
• Create a river coverage of straight and diagonal segments. Use the streamline algorithm, considering distance-to-mouth.
• Trace from source nodes to mouth(s), thereby pruning fluff on the river coverage. The resulting river can (if desired) be gridded and vectorized without undergoing changes.
• Intersect streams with contours. Flag all zero-slope and neg-slope segments.
• (if necessary) Manual edit. Flat segments are streams crossing contours. Redo the the stream intersections.
• TO BE CONTINUED NEW STUFF
• I found that I made a terrible assumption above. It is not possible to have convertable line and grid representations of a river. A diagonal segment will sometimes capture one of the cells that it touches only on the corner. The required fix is to arcpoint the river, then pointgrid it. This is tacked on to the end of my new rivgrid.aml.
• I have an improved aml for rasterizing rivers. After counting the number of subcells in each potential river cell, it uses that count to weight the direction of river flow. It creates a river coverage in which each vertex lies on a lattice point. Then it also recreates a mask grid of cells which match the coverage. (See rivgrid.aml above.)
• I have an improved java class for calculating channel slopes. (See above)
• I hope soon to be tracing uphill.
• I hope to update an integrated package soon.

---

Tackling the problem again in 2001

The task of calculating stream slopes varies with the data source. Some areas enjoy high-resolution laser altimetry (LIDAR) DEMs made from millions of spot elevations. It ought to be possible to map stream channels of unprecedented accuracy, and to calculate stream slopes on a pixel-pixel basis with LIDAR, though this has not been investigated here.

For much of western Washington, the best source of elevation is 10-meter DEMs. DLG 7.5' contours are slightly preferable, but coverage is sparse. Given 10-meter DEMs, 40-foot contours can be reconstructed to nearly duplicate the original contours. However, the algorithm that interpolates between contours is not as smart as you are. Therefore, the DEMs should not be trusted for stream slopes, or for the creation of contours (such as 10-meter intervals) that do not appear on the paper quad sheets. The matter is further complicated if the DEM has been built from a mix of quad sheets, some containing 20-foot contours, and some containing supplementary contours only at lower elevations. It can take some effort to reconstruct contours for regions spanning many quad sheets. The task is exacerbated by the processing and storage requirements of large datasets. Note,

• DEMs are generally metric, your 40-foot foot contours are 12.192 meter (or 121.92 decimeter) intervals
• Keep track of your projection and datum. The USGS DEM files for Washington are the same projection as the paper maps: UTM, NAD27, NAVD1929.
• For best results, create contours in the DEM's projection, then project the lines into your river projection, if required.
• In ARC/INFO, the GRID "contour" function is more powerful than the "latticecontour" command the many of us grew up with.
• If 10-meter DEMs are not available for your area, 30-meter DEMs are a poor substitute. If they are level-two DEMs, consider giving up.

Mapped rivers are generally superior to calculated rivers. In low-slope areas, calculated rivers are garbage. Sources of mapped rivers include (get out your checkbook) DNR, USGS 100K rivers, some USGS 7.5' DLG files, and various agencies.

The crux of the technique is to intersect the rivers and the contour lines. The best assumption is that the river slope is uniform between contour lines.

The first problem is making sure that all the rivers point downhill. In cases where the rivers branch in a tree-like structure, ARC/INFO provides techniques for flipping rivers. Where rivers are braided or form distributory patterns, hand-editing is necessary. Where drainage canals crisscross the landscape, it's anybody's guess.

Another problem is when streams do not align with the contour coverage. Streams can cross contour lines so that segments appear to be perfectly flat or to travel uphill. You will always get better results if these are fixed by hand-editing. When software encounters these problems, it attempts to ignore these bogus intersections. Often, though, the software flags river segments are susp... [ I will have to finish this later. In the meantime, you can look at my latest attempts, including new versions of amls, at sno2001].