Index of /areas/dryvalleys

[ICO]NameLast modifiedSizeDescription

[PARENTDIR]Parent Directory   -  
[IMG]fuzzytest.tif 2018-10-05 15:32 131M 
[   ]fuzzytest.lyr 2018-10-05 15:34 12K 
[TXT]READ_ME.txt 2018-10-05 15:35 355  
[IMG]southfill_1m.tif 2018-10-05 15:38 1.2G 
[IMG]testacc.tif 2018-11-15 15:55 102M 
[IMG]southslopem1acc.tif 2018-11-16 15:15 102M 
[IMG]lidar.png 2018-11-16 15:47 70K 
[IMG]fig_lidar.png 2018-11-28 16:13 97K 
[IMG]fig_lidar_smooth.png 2018-11-29 13:26 65K 
[IMG]fig_mosaic.png 2018-11-29 14:07 377K 
[IMG]acc_cm.tif 2018-12-09 14:30 109M 
[IMG]southslope.png 2018-12-10 12:30 57K 
[IMG]transects.png 2018-12-10 12:44 42K 
[IMG]acc.png 2018-12-10 14:38 583K 
[IMG]fig_acc_tryptich.png 2018-12-10 15:45 534K 
[IMG]hillshade_bumped.tif 2018-12-11 14:00 9.4M 
[IMG]fig_acc_melded.png 2018-12-11 14:17 865K 
[   ]z.tfw 2018-12-12 15:59 88  
[TXT]z.aux.xml 2018-12-12 15:59 2.5K 
[IMG]z.ovr 2018-12-12 15:59 25M 
[IMG]z.tif 2018-12-12 15:59 86M 
[TXT]z.xml 2018-12-12 15:59 5.8K 
[IMG]fuzzysouth.tif 2018-12-13 15:48 1.4G 
[IMG]fig_acc_3algo.png 2018-12-19 16:19 622K 
[IMG]april_acc_melded.png 2019-04-29 13:54 9.2M 

Dry Valley Hillslope Flowaccumulation

Dry Valley Hillslope Flowaccumulation

Prepared by Harvey Greenberg with Ping_chun Lin

This is the bare beginning of a web page. It will be filled in with more images, hyperlinks, and discussion.

The primary work directory is e:/dryvalleys on dusty in the puppetlab (JHN376). Some files were saved on the group fileserver at y:/topog/areas/dryvalleys. This page is on the web server w:/areas/dryvalleys


LiDAR Coverage. (from fig_lidar.mxd) A close view of the shaded relief shows odd north-south stripes which the data supplier could not explain. (centered on coordinates -17000,47200.)
To reduce artifacts, we tried smoothing over a one-by-eleven window, but the result showed strong degradation without total elimination of artifacts. As a compromise, we smoothed over a one-by-three window, and then aggregated the a 2-meter DEM, using mean values.
In the figure above, the bright rainbow streak shows the available lidar. Dark splotches are no-data areas where water or ice gave poor lidar returns.
The pinkish rectangle is the primary area for study. We explored the 2-meter DEMs created from pairs on satellite image swaths. (See y:/topog/world/REMA and http://data.pgc.umn.edu/elev/dem/setsm/REMA/indexes/ ) In accordance with the creators:

Acknowledging PGC Services (including data access)
Geospatial support for this work provided by the Polar Geospatial Center under NSF-OPP awards 1043681 and 1559691.

Acknowledging DEMs created from the ArcticDEM Project
DEMs provided by the Polar Geospatial Center under NSF-OPP awards 1043681, 1559691, and 1542736.

Acknowledging DEMs created from the REMA Project
DEMs provided by the Byrd Polar and Climate Research Center and the Polar Geospatial Center under NSF-OPP awards 1543501, 1810976, 1542736, 1559691, 1043681, 1541332, 0753663, 1548562, 1238993 and NASA award NNX10AN61G. Computer time provided through a Blue Waters Innovation Initiative. DEMs produced using data from DigitalGlobe, Inc.

Their footprints are represented on the image by gray lines. There are 187585 swaths for Antarctica. 53 swaths overlapped our area of interest. They were badly matched (leading to artifacts at swath boundaries, and had many no-data areas. We decided not to use them.

We chose instead to use the 8-meter composite tile built by REMA from the strip data. This data was reprojected and resampled to 2 meters.

In a 3600-by 7050 window, the lidar DEM was laid over the Satellite DEM, preserving the no-data lake in the lidar DEM. (It was a big mistake to try it without preserving the sink. The results of this mistake are preserved as southslope_nonodata.tif and southslopetoomuchfill.tif.) Six transects show the difference between the aerial lidar and satellite images where they meet. In happens that all six transects show the aerial images being lower, but and examination of accumulated area will show a few places where the aerial lidar is higher.
Flow accumulation methods
There are various methods of calculating flows (of water or whatever) across the landscape. The simplest method is to assume that all the water in a cell flows to its lowest neighbor. (Things are slightly more complicated when there are closed depression which are do not drain into noData cells, or when there are flat areas as can occur by chance or when closed depressions are filled by pre-processing the DEM.) This works well for streams (if you don't mind their being one cell wide, and having no braiding or deltas), but works badly for hillsides. ESRI offers an "MDF" flow flow direction option, and can use MFD files to drive the flowdirection function. However, ...
In order to have the minimum size undrained sink at the bottom of the valley, we expanded the noData area. After a few iterations ("more_sinks" and "sinks641" in e:/dryvalleys, we settled on "sinks6417", with all cells below 64.17 meters being redefined as noData. As it turned out, all this effort made little difference.

We filled all the sinks in the default ESRI manner, and calculated flowaccumulation with shalstab.py The black box indicates the part of the image which is blown up on the right. The blue fish-shaped area near the top is an area where the shalstab.py algorithm breaks down. It is not important to this analysis, but we are looking into the problem.

In the lower box of left-hand figure, we see a region of fuzzy parallel flow lines. This is because sinks have been filled, and that region is represented as flat when flows are calculated. In the second figure, we filled only those sinks less than one meter in depth. Flows in that are represented better, but total flows throughout the landscape are diminished as flows enter sinks and stop. In the third figures, there is even less sink filling. Note the fish-shaped figure in the top box. For reasons we have not determined, this artifact is reduced when there is less sink filling.
Our ultimate flow-accumulation product melds the three versions, using the default full-sink-filling version for most of the landscape, but substituting data from the other two models where the first model does not represent the flows well. (The underlying shaded-relief image was made with "resolution bumping", where the DEM was averaged with a smoothed DEM before hillshading. This emphasizes general forms without completely obscuring fine details.


Addendum on flowdirection/flowaccumulation

Many of the tools that we have built by hand have since been supplanted/supplemented by tools within GIS packages such as ArcMap. The traditional method of calculating flow direction has been the D8 method, where all flow goes in the steepest direction to one of the source cell's eight neighbors. Now (in version 10.6.1 of arcmap) there are two other options. The first is the D-Infinity model, which yields an angle between 0 and 360 degrees. On a smooth landscape, the results are close to the results of the "Aspect" function, though they are confusingly presented as cartesian angles (counterclockwise from eastward=0) instead of bearings (clockwise from northward=0). Though this suggests the thinking behind the DEMON program, it does not lead to flowaccumulation in the ESRI environment. The third option is Muliple Flow Direction (MFD), which records all lower cells. The result is a .crf file with a pixel depth of 80 bits, and a poorly-documented structure. ESRI has no way of displaying the data in this file, except for the D8 direction. However, it can be used as input for flowaccumulation. Compare the result (middle panel) to the other algorithms:
three methods. A simple example shows how strange the ESRI algorithm is.