Satellite Remote Sensing:Flood Damage Assessment in Pacific Northwest
After recent widespread flooding in the Pacific Northwest, enhanced satellite imagery is offering a unique look at landslide damage.
By Ruth Carapella

In early 1996, an unusual combination of weather events in the Pacific Northwest created both a catastrophic flood and the opportunity to obtain clear satellite images of the flood's peak. Two weeks after the sensors captured the images, managers at the USDA Natural Resource Conservation Service (NRCS) began using images of the south half of the Willamette Valley for flood damage assessment work. Atterbury Consultants began testing a colorization process on the north half of the Willamette drainage. The goal of the enhancement was to help the Forest Service detect landslides.

The Event and the Images
In mid-January, with the snowpack measuring only 29 percent of average, the chances of a flood in the Willamette Valley of western Oregon seemed remote. Over the next two weeks though, extreme winter storms made up the shortfall. By the end of January, the average snowpack for the Willamette drainage increased to 112 percent of average.
      On Feb. 6, a strong subtropical jet stream reached Oregon. This warm, humid air mass brought record amounts of rain. Additionally, the warm subtropical jet pushed freezing levels upward, causing a rapid melt of the snowpack. Streams and rivers throughout the region rose quickly.
      Bracing for widespread flood damage, Mon Yee, natural resource manager with the NRCS, contacted Robert Wright, information systems forester at Atterbury Consultants Inc. Yee asked Wright to conduct an image search for relevant coverage from several satellite image companies.
      Wright was already at work on the image search. He had learned that both the Landsat and SPOT sensors would pass over the Willamette Valley on Feb. 10 and 11. Wright contacted EOSAT and SPOT Image Corp. and expressed Atterbury's interested in obtaining images on those days. It was a long shot. Clouds had blanketed the area for days. A glint of sunlight peaked through the clouds Friday, Feb. 9, and Wright hoped for the best. But Saturday, as the satellites covered the area, neither satellite could get a clear view.
      Friday, Feb. 9, 7:00 p.m., the Willamette River crested at 28.6 feet in Portland, 10.6 feet above flood stage. Hydrologists estimated a flow of 365,000 cubic feet per second. Sunday, Feb. 11, the sky cleared. Within one half hour of each other, and 41 hours after the Willamette's crest in Portland, both Landsat and SPOT captured crystal clear images of the Willamette Valley.
      The quality of the images and the timing of their acquisition could not have been better. Wright said, "This is the first opportunity in the Northwest to have such stunning clear imagery so close to the peak day of a flood. There's not even a jet contrail."

Using the Images
Atterbury prepared before- and during-flood Landsat images of the south half of the Willamette Valley as a visual reference for the NRCS. Atterbury used a 1994 Landsat image from their database for the before-flood image.
      The pair of images attests to the volume of water and magnitude of flooding. With flood widths greater than that of the Columbia River, the normally small Pudding and South Yamhill Rivers make a striking appearance. Downstream from Salem, the Willamette River spread to a width of four miles. Vancouver Lake, normally separated from the Columbia River by a stretch of land, appears only as a dark blue pool within the river itself.
      Yee said, "The images provide a broad synoptic view that other media cannot. This broad view helps people comprehend the magnitude of the event. We are using the images to evaluate the extent of the flood, especially at the crest cycle."
      Eventually, Yee plans to incorporate Landsat digital data as a layer in the NRCS GIS and as a tool for evaluating previous flood plain delineations. Landsat data will be evaluated with soils and geology information to assess watershed health. Within each watershed, the images will help the agency establish priorities for watershed restoration work.
      Assessing flood damage poses a challenge to mangers in the Pacific Northwest Region of the USDA Forest Service. Forest Service hydrologists, biologists, and engineers are interested in locating landslides and areas of mass erosion in the back country before additional snow melt and rain compound damages. The agency must assess the damage to develop cost estimates for restoration work.
      Typically, the Forest Service's first step in damage assessment would be to obtain aerial photos at 1:24,000 or larger scales. However, as Chris Fayad, remote sensing specialist at the Regional Office in Portland said, "In this case, the satellite images were acquired at an optimum time: right after the peak of the flooding and right after many of the landslides occurred."
      Additionally, the weather did not cooperate for obtaining timely aerial photographs or videography. Some areas were flown, but winter clouds returned and storms blanketed the hillsides with snow, hiding landslide scars.
      The Forest Service has not decided if satellite images will provide adequate information for landslide detection work; but, Fayad believes the images can serve as a tool for providing interim information to land managers. Fayad said, "You can get a rough estimate of damage. You can see the drainages with massive slides and this can give good quantity and damage estimates for determining the amount of money needed for repair."
      There is little precedence for using satellite images to detect such widespread landslide damage, so the Forest Service was interested when Atterbury extended an offer to try an enhancement process that might aid detection. The goal was to increase the number of detectable slides. Since small individual landslides can be hidden by vegetation, no image enhancement process would allow detection of all slides.

Colorizing the SPOT Image
Using satellite images to assess flood damage is not a new idea. In the Midwest floods of 1993, the Federal Emergency Management Agency (FEMA) required use of Landsat data for flood damage assessment work. SPOT and ERS-1 radar provided collateral sources of information.
      EarthSat, who completed most of the 1993 Midwest flood analysis for FEMA, cited the advantages of using satellite images instead of aerial photographs on this project. Satellite images provided a synoptic view at a lower cost per unit of land, multispectral sensitivity allowed detection of flood extent even after waters receded, and the data collected could be analyzed in a digital format. The chief disadvantage of satellite images was the coarse spatial resolution of the data.
      Atterbury hoped to diminish this disadvantage by combining the spectral information from the Landsat image with the higher resolution SPOT image. To do this, Jeff Jenkins, senior project manager at Atterbury Consultants colorized the SPOT image.
      Jenkins used TNTmips software and the during-flood Landsat image with a 30 meter resolution and a 25 meter pixel size.
      In TNTmips, Jenkins identified the TM file for resampling to the 10 meter SPOT resolution. Then, using the software's nearest neighbor method, he resampled the image. To run the automatic resampling process in TNTmips, he set the model to affine, the scale to match the reference scale, the orientation to match the projection, and the extents to match the reference. Processing took approximately 10 minutes on a 133MHz Pentium PC with 32MB RAM.
      Next, Jenkins displayed the multiple raster layers of the TM scene with the SPOT panchromatic band. Jenkins said, "The TM scene is resampled to the SPOT scene. During display, the software adds the gray scale values of the SPOT panchromatic band back into the TM scene. The software interprets the gray scale value for each cell as either a brightness value or an intensity value. The user determines which by clicking on either the RGB-B or RGB-I option. The higher the gray scale value of a particular cell in the SPOT scene, the higher the brightness or intensity of that cell in the TM multispectral band."
      Jenkins experimented with the colorization process. He generated one image using the three natural color bands and another using bands 5, 4 and 3 (middle infrared, near infrared, and red band widths). Though both colorized images achieved the objective of adding spectral information to the SPOT image without loss of resolution, Jenkins doubted either product offered the best resolution for landslide detection.
      Fayad concurred. "I don't know that it adds a lot of information. Using the IR band created nonsense data in the shadows, probably because the IR is absorbed in the shadows and the shadows are extremely dark." Fayad also pointed out that the colorized SPOT image contains no more spectral resolution than the Landsat image by itself. The resampling process breaks each single pixel into several pixels, all with the same spectral value.
      Because of the loss of information in the shadows, these areas appear as white pixels on the colorized image. This problem does not occur on the image colorized with natural colors; but, the natural colors do not produce dramatic spectral signatures either.
      From a cursory review of the colorized image, Fayad and Jenkins concluded that landslide detection may be based more upon spectral differences than textural differences. Consequently, this type of artificial enhancement may not facilitate interpretation. Fayad thinks the agency could gather adequate information for initial planning work from the TM scene alone if the Forest Service decides to pursue use of the satellite images.
      Future Analysis Although both Fayad and Jenkins agree that these artificially colored SPOT images may not enhance landslide detection, both believe the colorized SPOT image may aid interpretation for other resources.
      Jenkins said, "I think there is room for using one, two, or three TM bands in different combinations. The user can adjust the contrast of the TM bands using the SPOT image or by using other software tools such as filters to isolate reflectance signatures."
      Having gone through this process may help the Forest Service choose the best tools for landslide detection later. Jenkins said, "Atterbury reached into their tool box to show some potential of the images. When it comes to image interpretation, sometimes you just have to plant both feet and punch it out. There isn't a lot of precedence in using satellite images for widespread landslide detection. Sometimes you have to try a few different approaches before you find the best combination of tools to get the results you need."
      Jenkins pointed out that the colorization process may achieve excellent results when applied to a different image base. For example, aerial photos could be colorized with information from the Landsat image bands.
      This approach was very successful on a recent Atterbury project in Argentina. On that project, Atterbury created a black and white photo mosaic and then colorized that photo mosaic with three bands from a SPOT multispectral scene. Slight textural differences on the photos and slight spectral differences on the Landsat image complemented each other on the colorized image, creating dramatic differences between two key timber types.
      As agencies collect more data, more opportunities for image enhancement and analysis will develop. Currently, several agencies are obtaining aerial photographs. The Corps of Engineers contracted for GPS controlled videography. The NRCS contracted for color IR slides of the Willamette's tributaries in the Willamette Valley and major drainages in Tillamook and Clatsop Counties. The Forest Service contracted for aerial photography over the Siuslaw and Mt. Hood National Forests and in the Columbia River Gorge National Scenic Area. Some areas have been flown; others will be flown soon.
      Fortunately, managers don't need to wait for that data to start making plans for damage repair work. Thanks to the short window of sunshine that allowed the capture of these excellent Landsat and SPOT images and Atterbury's efforts to make the images available quickly, managers from the NRCS and the Forest Service can start planning now.

About the Author:
Ruth Carapella writes about forestry and natural resources for Pen Craft Writing & Editing Services in Harrison, Idaho. She may be reached at 800-689-9235 by phone or fax.

Back