Spatial Analyst to create streams from LIDAR GRID

[ChristopherJursa]

Hello,

I am attempting to create a series of streams from a 1 meter resolution LIDAR DEM in GRID format.

I am using ModelBuilder to create a workflow.  Basically, I do the following.

Run a fill
Run a low filter
Generate flow direction
Generate flow accumulation
Conditionally use only those with 100 cells of accumulation
Create stream links
Create stream order
Convert streams to features

I have attached my workflow (workflow.jpg).  It executes and outputs streams.  However, the streams are fragmented and appear as dashed lines that don't really connect. 

My map attachment (map.jpg) shows this.  Why does this fragmentation of the streams occur?  I want my new streams (in green dashes) to look something like the original streams (in red). 

I noticed if I run an aggregate (of 3 or 9 meters resolution cells), I get better connected streams.  Why won't this work with 1 meter resolution cells?

Again, why are the streams so fragmented?

Also, is my workflow correct?  Am I using the stream link, order and features operations correctly?  Should I even run a filter?

Thanks in advance!

[Anonymous User]

Original User: jevans02

Looking at your results it seems that several things may be influencing your results. It is possible that the lidar filtering algorithm did a poor job at identifying ground-only returns. Sometimes, regardless of how well the ground filtering algorithm performs, the resulting DEM is highly variable. You most likely need to smooth the DEM using a FOCALMEAN or resample the DEM. You do not need a 1m DEM to delineate a stream network. I would also look back at the interpolation that created the DEM. Is the resolution of the DEM supported by the resulting point spacing of the filtered data (if not artifacts can be introduced)? What interpolation approach was used? In my experience IDW and block minimum are a very poor choices. In ESRI products the best results are from the TOPOGRID model (Spatial Analyst Tools > Interpolation > Topo to Raster). When using lidar derived DEM's one needs to think about what ground is. In reality ground is highly variable (just take a walk outside) and lidar is directly sampling this variability. DEM's used for hydrological modeling need to be quite smooth so water is not routed in random directions. Standard tools for hydrological conditioning of DEM's are not designed to address this type of variability. When using extremely high resolution DEM's it is necessary to evaluate the data and then apply any necessary preprocessing before even starting watershed/stream delineation.

For alternative tools for watershed/stream delineation please see my reply to this thread:
http://forums.arcgis.com/threads/4797-Spatial-Analyst-how-to-model-rivers

[ChristopherJursa]

I have attached an updated workflow.

I new use an aggregate immediately after the filter.  Should I use the focal mean instead?

I did fix the problem with the parsed lines.  Strangely enough, when I do the steams to features in the workflow, I get the parsed lines.  However, when I run the function separately in the toolbox, it delivers correct results.  Why is that?

The algorithm still seems to be missing some streams when I compare it to my hillshades.  Any ideas why? 

I have attached am image where my streams are blue and the previously issues streams are in red.  The red streams are correct in this case as this is a river.  Some of my blue streams do not connect as well.  Why does this occur?

Finally, I am creating the streams from my LIDAR DEM GRID which was originally in State Plane PA North 1983 in feet.  I reprojected them to UTM Zone 17 1983 in meters.  As far as the elevation values, I converted those from feet to meters manually and they are correct.  Could the reprojection of the cells be causing problems with ArcGIS recognizing stream features?  I noticed a blocky look when zoomed into the UTM projected GRID.   Just a thought.

Thanks for your previous help!

[Anonymous User]

Original User: jakubsisak

I have attached an updated workflow.

I new use an aggregate immediately after the filter.  Should I use the focal mean instead?

I did fix the problem with the parsed lines.  Strangely enough, when I do the steams to features in the workflow, I get the parsed lines.  However, when I run the function separately in the toolbox, it delivers correct results.  Why is that?

The algorithm still seems to be missing some streams when I compare it to my hillshades.  Any ideas why? 

I have attached am image where my streams are blue and the previously issues streams are in red.  The red streams are correct in this case as this is a river.  Some of my blue streams do not connect as well.  Why does this occur?

Finally, I am creating the streams from my LIDAR DEM GRID which was originally in State Plane PA North 1983 in feet.  I reprojected them to UTM Zone 17 1983 in meters.  As far as the elevation values, I converted those from feet to meters manually and they are correct.  Could the reprojection of the cells be causing problems with ArcGIS recognizing stream features?  I noticed a blocky look when zoomed into the UTM projected GRID.   Just a thought.

Thanks for your previous help!

This is a strange hillshade. It looks like it was created from a raster that was converted from a TIN. (judging by the triangles.)  If you are running your analysis using this data as you input, it might be the cause of your problems. Make sure you are using the raster that was created from the original XYZ data.
Also, generalize the raster; 1m resolution is a bit of an overkill.  If your input raster is a floating point raster, convert it to integer raster before you do anything.

[JeffreyEvans]

You are adding artifacts by aggregating the DEM. Aggregation is not intended for continuous data. You should smooth and resample the data. Use the bilinear or cubic option when resampling.  I believe that it would be a mistake to convert the DEM to integer. Truncating the float values tends to bring out interpolation bias in the data. Seeing the hillshade the original source data is VERY noisy and you will not get satisfactory results until you deal with this variability. I have posted advice to you in a previous post. Please read and take note.

[ArthurCrawford]

Another option would be the tools from the Center for Research in Water Resources.   I have used their Optimized Pit Removal to quickly process bad data to then get streams for large areas.    This tool is good for getting streams in areas where you have lidar.

Link: tools.crwr 

Optimized Pit Removal

Contains ArcGIS (version 10.1) geoprocessing script tools for removing pits (or sinks) from Digital Elevation Models using a combination of cut and fill. This alternative to the standard Fill tool provides more realistic flow paths with less required manual adjustment. Ideal for high-resolution datasets such as LiDAR.

[LornaMurison]

Has anyone used this tool with ArcGIS 10.2?

[curtvprice · Accepted solution]

This workflow is incorrect. Any smoothing of the DEM must be done before the Fill operation, and Flow Direction should be run directly after Fill.