Satellite image classification of the White River's surface near Rangely, Colorado

Rick Moran - May 2008

ES 775, Advanced Image Processing, Emporia State University


Satellite imagery of sun glint off a river was used to classify a small river's surface texture.   A 17 km portion of the White River located in Rio Blanco County, Colorado, USA was studied.   The results imply that useful information was obtained from sun glint, but the level of confidence in interpreting that glint is uncertain.

Ikonos satellite

Satellite imagery was obtained from GeoEye's Ikonos satellite.  The Ikonos satellite orbits the Earth every 98 minutes at a 680 km altitude.  The satellite can produce 1 meter imagery every 3 days over the same area.   Instrumentation includes a multispectral 4 meter and a 1 meter panchromatic sensors. The panchromatic sensor covers 0.45 to 0.9 mm.  The multispectral bands are as follows:  band 1 is 0.45 to 0.52 mm; band 2 is 0.51 to 0.6 mm; band 3 is 0.63 to 0.7 mm; band 4 is 0.76 to 0.85 mm.  Imaged pixels are referenced into standard coordinate systems.  The company provides a straight forward process at their website to select archived imagery for purchase.   Geographic coordinates are entered into an interactive application from which low resolution images are viewed before acquiring the detailed images.   Alternatively new images can be ordered on demand.  From GeoEye imagery was acquired from the data archive.   The August 2006 images provided were the multispectral resample at 1 meter from the panchromatic imagery.  The imagery presented here represents a tiny fraction of that dataset.

White River

The White River with headwaters in eastern Rio Blanco County, Colorado, USA flows predominately westward 200 miles to the Green River in Utah.   The Green River merges further to the south with the Colorado River.   The White River was studied where it passes through Rangely, Colorado.    Through Rangely the river is 25 to 40 meters wide with banks along the side typically 1 to 3 meters high.  A map of the river is available from TerraServer .

A 3 km section of the river was canoed to obtain basic measurements of water depth, current velocities, and record descriptions of surface texture at multiple locations.   It soon became apparent obtaining measurements at stationary points was challenging for 1 person in a canoe due to the current.   Using an anchor proved to be impractical since the canoe became unstable in faster currents.   Several river cross sections were recorded.   Typically the river has a maximum depth of 2 meters with some cross sections having a maximum depth of <1 meter.   The deepest location measured was 3.5 meters and is known as a good spot for cat fishing.  More dynamic water surface texture was typically associated with depths < 1 meter.  The measured current was commonly 1.5 km / hour with the strongest currents 4 km / hour.   The surface texture various from near mirror to 0.15 meter tall ripples to frothy.

The water surface texture was noted to correlate only slightly with water depth and velocity.   The surface texture typically is determined by the shape and depth variation in the river channel 10 to 50 meters upstream.  Visually the amount of suspended sediments appears consistent along the stretch of river observed.  The entire length of the river imaged is underlined by the Mancos shale which gives the river a muddy consistency.   The water bottom cannot be seen beyond a depth of 0.2 meters.

  White River passing through Rangely.   Most of the river's banks has vegetation 0.5 to 3 meters in height.   Cottonwood trees are occasional found a few meters from the river edge.


This section of the river has a high level of sun glint associated with larger ripples


A small land feature in the middle of the river.   The glint on the right of the photo is most typically associated with smaller more random surface texture.


  5 to 20 meters down river from where the river is significantly agitated a frothy foam often exists that will break down with 50 meters down stream.   A cluster was identified that has a high component of foam.


To determine what river only clusters could be extracted the following processing was done using Clark Labs -IDRISI's software:

1) To correct for atmospheric conditions, 1 was added to each band using the SCALAR module.   At an altitude of 5300 feet and low humidity the haze was minimal.

2) The WINDOW module was used to reduce the image to an area covering the river.

3)  An unsupervised classification using ISOCLUST was run with 60 clusters and all 4 bands.

4) To develop a mask the RECLASS module used cluster 6 to assigned a value of 1 and the remaining clusters 0.

5) The GROUP module was run creating 311,042 groups.

6) Due to bridges crossing the river it has divided into 3 groups.   A RECLASS was performed which assigned the 3 group segments to a value of 1 and the remaining groups 0.

7) Each band was multiplied by the river mask with the OVERLAY module to produce an image of just the river.

8) A median 3x3 FILTER was applied to each band to smooth the image.

9) An ISOCLUST with 10 clusters was used to produce an unsupervised classification of just the river.

A detailed section of the river with unsupervised classifications.  Most of the short narrow channels off the side of the river provide livestock access to river without having to step into the rivers current. The majority of the river is grouped in clusters shown in blues.

The unclassified clusters use to isolate the water bodies along with other features listed in processing step #3.

The groups created in processing step #5.  The feed was the green cluster in the previous image.

Scatter plot of band 1 on the y-axis and band 2 on the x-axis.   Most of the clustering information is contributed from bands 1 & 2.


Defining real world meaning to the unsupervised classification is problematic.   Sediment suspension is sufficiently uniform that it can not be used to explain the clusters.   Water velocity showed no direct correlation with any cluster.   It is likely some clustering is influenced by reflected vegetation of the water's surface in particular when river banks are high or cottonwood trees are present.   Small format aerial photography would  be useful to provide photo's from nadir and varied offset angles to clarify any reflected vegetation influence.  The field ground truth was done 2 years after the image was acquired when the water flow was somewhat higher.  But the difference in flow rates didn't appear to impact the waster surface texture significantly.  The assumption was made that wind was a minor component based on observing the river from the shore during a moderate wind.

Most all of the cluster segmentation is likely due to sun glint.  The glint reviewed here is likely dominated by surface texture variations due to river flow turbulence.  Typically efforts are made in processing to remove sun glint.  Some sophisticated algorithms are used to accomplish that goal.  Taking advantage of sun glint has been suggested and some effort has been done to investigate ocean waves.

A best effort was made to identify the clusters as meaningful water surface textures.   The clusters don't necessarily match what I observed in the river to the detail the clusters would indicate.   Relative changes in surface texture match better to clusters when the river is orientated east to west.   The surface texture variation indicated by the clusters generally is understated for river flow to the south or north.

From the ~60 water depth measurements there was a correlation to 1 cluster with water depth.  The green cluster identified as shadows has the highest level of confidence in correlating it to a real physical characteristic.   This cluster is almost always on the south flank of river with taller banks.  That is the river bank casts a shadow on the river.  A somewhat high confidence exists for clusters identified as frothy and shallow.  The frothy clusters typically are associated with white foam on the surface generated from a surface disturbance upstream.   The level of confidence is very low on the brown shadows and may more properly be identified as unknown.

A 17 km section of the White River processed with unsupervised classification


This western section of the image has the most variation in clusters.   It appears to result from the varying river channel, but since the image is slightly more off nadir the sun glint may have increased.


A effort was made to identify the spectral variation occurring in the satellite image of the White River.  The variation appears dominated by sun glint that is associated with water surface texture changes.   Textual differences can be grouped into clusters that appear to represent long term water surface disturbance, but caution is needed not to read to much into the classifications.    There will remain the uncertainty in attempting to interpret sun glint.


Dinosaur Diamond Prehistoric Highway. Canoeing the White River

Gilvear D., Davids, C., Tyler, A., 2004. The use of remotely sensed data to detect channel hydromorphology; River Tummel, Scotland

Irving, D., Haines, B., Modde, T., 2003. White River Base Flow Study, Colorado and Utah, 1995-1996

Montagner F., Billat V., Belanger S., 2003. Sun Glint Flag Algorithum

Mountain Wayfarer, 2004.  White River Run - Canoe and Kayak Race, Rangely, Colorado

National Weather Service. Advanced Hydrologic Prediction Service: Grand Junction: White River

Parsons, K., Parsons S., 2003.  A Boating Guide to the White River

River Runners Transport, 2001. White River -- Colorado and Utah River Running

Robinson, I., 1983. Satellite observations of ocean colour

Tatham, B., O'Brien, D., 2001. Bringing Raster GIS to The District of North Vancouver

Veenstra, T., 2006. Polarization-Enhanced Sun-Glint Imaging for Measuring Ocean Wave Statistics

Wang, M., Bailey S., 2001. Correction of Sun glint Contamination on the SeaWiFS Ocean and Atmosphere Products

Wikipedia. White River (Utah) - Wikipedia