Hurricane Floyd: Sedimental Reasons
by David Herring
October 2, 2000
  Page 2

River ecosystems are dynamic; they profoundly shape and reshape their environments on an ongoing basis. Pick any river on a map and a biologist can tell you about the myriad plant and animal species that live there. Likewise, a geologist can tell you how the river etched away at its riverbed, continually carrying tiny bits of it downhill and, over millennia, helped carve the surface into the shape it has today. While biologists study how riverine life adapts to the ever-changing chemical and physical state of the ecosystem, geologists examine the processes by which rivers gradually transport vast amounts of minerals and sediment downstream, thereby changing the landscape.

A new science discipline emerged recently, and is evolving, that focuses on the "biogeochemical cycles" of rivers and other water bodies. As the term suggests, scientists in this discipline study how changes in the chemical and geological state of a body of water affects plants and animals living there, and how life, in turn, influences the chemical and geological makeup of the ecosystem.

"The transport of water and sediment through wetlands and floodplains influences not only the rates and types of biological and chemical processes, but also the geomorphic processes constructing the landforms," explains Leal Mertes. She is an associate professor of geography at the University of California at Santa Barbara and a co-investigator on the NASA-supported Global River Flood Studies team (part of the Earth Observing System program).

Mertes helped develop and refine a technique for using satellite remote sensing data to estimate the amount of sediment suspended in surface waters of rivers and wetlands (Mertes et al. 1993; Gomez et al. 1995). Her technique is useful for scientists who want to assess the impacts of increased sediment runoff during a flood event, and how this geochemical change in the water may impact the life forms living there.

"Satellites are particularly useful if the river is somewhere geographically inaccessible," Mertes explains. It is not always practical to travel to flood sites wherever and whenever they occur—sometimes in the heart of the Amazon rainforest or other remote regions where it is difficult to transport gear in time to measure the event. Yet on any given day there is a variety of Earth observing satellites that will see every region on the Earth's surface, providing scientists with invaluable data about floods.

Mertes and her colleagues within the Dartmouth Flood Observatory (G.R. Brakenridge and E. Anderson) comb news media all over the world and look for reports of floods. When they find one, they begin poring through satellite data archives to find out if any was able to produce images of the event. Then, once they obtain the data, they go to work measuring the extent of the flood, mapping and remapping floodplains (often in places where no maps previously existed or were inaccurate), and measuring sediment concentrations in the water.

Mertes and her colleagues were particularly interested in studying the severe flooding in North Carolina brought by Hurricane Floyd in September of 1999. News reports painted a bleak picture of highly-polluted rivers carrying human and animal sewage, topsoil, pesticides and fertilizer, and even animal carcasses in the runoff. Satellite images showed a large plume of sediment spewing out of the rivers and spreading out to sea off the Carolina coast. Consequently, state officials predicted there would be a massive fish kill.

 
 

Hurricane Floyd Series:
Hurricane Floyd's Lasting Legacy
Fearing the Worst
Sedimental Reasons

 

sediment along the Florida coast

This image by the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) provides a clear indication of the power of Hurricane Floyd. The waters along the Florida coast in this September 16, 1999, image appear deep blue except for an irregular narrow band along the coast which is colored yellowish-brown by suspended sediments and dissolved, colored organic materials (CDOM) that wash into the ocean from rivers and are resuspended by the action of the surf and currents in the shallow coastal waters. As Hurricane Floyd passed over these waters, he stirred and mixed them much deeper than normal—deep enough, in this case to resuspend sediments that had settled on the continental shelf nearly 200 meters below the surface. At the shelf edge the water becomes rapidly deeper and bottom sediment could no longer be reached by the hurricane. (Image provided by the SeaWiFS Project, NASA Goddard Space Flight Center, and ORBIMAGE.) click for enlargement

DEM model of North Carolina

Given the wide extent and the long duration of the floods, Hurricane Floyd gave Mertes a perfect opportunity to test her technique using new data from Landsat 7's Enhanced Thematic Mapper plus (ETM+) sensor, launched just months earlier. How much sediment was flowing through North Carolina's waterways? How extensive were the floods? How accurate were the floodplain maps there? Without even leaving her computer laboratory at UCSB, Mertes would find out.

next  Assessing Floyd's Floods

 

This digital elevation model of North Carolina shows the much higher-elevated Appalachian Mountains to the west (white and red pixels), the state's hilly piedmont region in the center (turquoise pixels), and the low-lying, relatively flat coastal plain to the east (dark blue). Notice the complex network of rivers running from the highlands toward the ocean. These waterways were etched into the face of the landscape over geological time by melting glaciers and rainwater runoff. ( Image by Brian Montgomery; data courtesy NIMA)


  Assessing Floyd's Floods Page 1Page 3

The Flood team obtained data from three different satellites to measure the extent of the floods in North Carolina and to analyze sediment concentrations in the water. Each of the satellites has certain advantages and disadvantages. The NOAA Advanced Very High Resolution Radiometer (AVHRR) has wide spatial coverage, enabling it to see most of the Earth's surface every day. However, its resolution is relatively coarse-each pixel represents one square kilometer-so it was difficult to glean detailed information about the flood from AVHRR. To compensate, they used Landsat 7 and Radarsat, both of which provide much higher spatial resolution (up to 15 meters per pixel), but they revisit a given point on the surface much less frequently (about once every 8 days over eastern North Carolina). Some rivers can actually overrun their banks and then recede to near normal conditions between two consecutive Landsat and Radarsat acquisitions.

 
 
Data comparison

This flood map shows a comparison of the flood data collected by AVHRR (pink pixels), Radarsat (red pixels), and Landsat 7 (green pixels). These colors show areas that were flooded in the wake of Hurricane Floyd. Due to interference from clouds and forest canopies, geographers often get the best insight into a flood when they use data from multiple satellite sensors. (Image courtesy Dartmouth Flood Observatory, E. Anderson and R.H. Brakenridge) click to enlarge

On September 17 and 18, 1999, AVHRR acquired clear shots of the floods, while Radarsat got its first look on September 23 and Landsat 7 obtained its first cloud-free image over eastern North Carolina on September 30. Because there were two major Hurricanes (Dennis and Floyd) within 11 days of one another, interspersed with heavy thundershowers, the ground was saturated and the floodwaters remained high during this entire period. Each satellite was able to capture valuable information about the ongoing event.

Mertes points out that measuring flood extent is not as simple as merely counting pixels of water in an image and then adding up the totals. Much of the flooded areas can be hidden under dense vegetation canopies where they are hard to see. Radar data are particularly effective for detecting water under vegetation (Alsdorf et al. 2000), but are not effective for detecting water color. In contrast, because of its high spatial resolution and its sensitivity to certain wavelengths (colors) of light, the ETM+ aboard Landsat 7 enabled Mertes to detect water quality (with respect to sediment) in flooded areas. Combining data from each of these different instruments provides the most complete picture.

Mertes explains that when sunlight hits water, some of it is immediately reflected by the surface back up into the air, and some of the light is refracted but penetrates beneath the surface, where some of it is absorbed and some is scattered by water and sediment particles. Some of the scattered light is reflected back up through the surface and up through the atmosphere. Satellite remote sensors can measure this "water-leaving radiance" (backscattered light). The more sediment particles there are in the water, the greater the amount of backscattered light. In short, whereas clear water appears dark in optical wavelengths because it tends to absorb most incoming sunlight, sediment-rich waters appear brighter because they reflect more sunlight.

Mertes uses a computer program to quickly scan an entire satellite scene looking for both the darkest and the brightest water pixels. Because these two pixels represent the extremes of high and low reflectance values, she labels them as "endmembers" and then plots the reflectance of each pixel at a range of visible and near infrared wavelengths (from 400 to 900 nanometers). The reflectance values for all other pixels containing water will fall somewhere between her two endmembers, but will somewhat resemble their unique shapes, or "spectral signatures" (Mertes 1997).

 
 
legend color for flooded area Sept 17-18

flooded area on Sept. 17 and 18 from AVHRR band 2 data.

legend color for flooded area Sept 23

flooded area on Sept. 23 from Radarsat Extended High 4 Beam data.

legend color for flooded area on Sept 30

flooded area on Sept. 30 from Landsat 7 Panchromatic band data.

endmember diagram

Next, Mertes begins to look for spectral signatures over regions that differ from her water endmembers. Those pixels with dramatically different reflectance curves can be identified as other surface features, such as land surface, concrete, or tree canopies. But those vegetated pixels with only slightly varying reflectance curves probably also contain some standing water. According to Mertes, the probability of there being water underlying the canopy is a function of how closely the reflectance curves match. Above a certain difference threshold, Mertes can flag those pixels that are purely vegetated surface and distinguish them from those that contain floodwater.

Using this technique on the three major rivers that overran their banks in eastern North Carolina—the Neuse, Tar, and Roanoke Rivers—Mertes determined that a total of at least 790 square kilometers were flooded. Of that combined area, 520 square kilometers were clearly open water, while at least 270 were flooded vegetation.

But could she tell how much sediment was contained in the runoff? Or, where the sediment came from?

next Floyd's Lasting Legacy
back Sedimental Reasons

 

This graph shows typical endmember measurements that Leal Mertes uses to determine the amount of sediment suspended near the surface in a given body of water. The top and bottom lines show reflectance values for three different samples at each of four different wavelengths that correspond to four of Landsat 7's bands. The top and bottom lines are based upon measurements made from samples in the laboratory, with suspended sediment concentrations of 5.6 milligrams per liter (low) and 207 mg/L (high), respectively. The middle line shows reflectance values collected by Landsat over a body of water that contains an intermediate concentration of suspended sediment. (Graph courtesy Leal Mertes)


endmember legend

  Floyd's Lasting Legacy Page 2

Unfortunately, due to heavy cloud cover over eastern North Carolina for an extended period of time in September 1999, high-resolution satellites like Landsat 7 weren't able to obtain clear views of the surface until well after Hurricane Floyd had passed.

"This was an interesting storm sequence," says Mertes. "You had three major storms go through the system in a relatively short time-first Dennis, then Floyd, then another tropical storm only days later. After Floyd, the system had already generated a lot of sediment discharge. It was interesting to observe which parts of the landscape continued to generate sediment with added flooding from the tropical storm and which parts didn't.

"Alternatively," she continues, "it was interesting to identify point sources of clear water. Where was the water running off the landscape cleanly? Sediment concentration maps can tell you that."

To determine how sediment was transported in surface waters, Mertes again examined how much light was reflected or absorbed by the water. She explains that the relationship between sediment concentration and backscattered light is not linear. As photons of light are reflected by sediment particles-often bouncing off many suspended particles like a pinball before being sent back up through the atmosphere-the length of the path that the photons travel increases, thus giving the water particles more opportunity to absorb more of the light. So, as scattering increases, the intensity of the light reflected back up through the atmosphere increases non-linearly because of competition for the light by absorption (Mertes et al. 1993).

Mertes observes that this effect varies at different wavelengths. She finds that the wavelengths between 400 nm and 1000 nm are the most effective for estimating sediment concentrations. Mertes has combined NASA data and her own laboratory data of reflectance measurements over water samples containing a variety of concentrations of different types of sediments to create a "catalog" of spectral signatures corresponding to sediment concentration. By cross comparing the spectral curves of the remote sensing data from the North Carolina floodwaters to the spectral curves of the laboratory samples, she can estimate the values of sediment concentrations in the rivers accurately to within plus or minus 20 milligrams per liter (Mertes et al. 1993).

By the time Landsat 7 was able to acquire a clear view of North Carolina (September 30), Floyd had passed and most of the floodwaters were coming from the tropical storm. Mertes found that, by this time, the waters in the flooded areas of the Neuse, Tar, and Roanoke Rivers were fairly clear. Most areas contained less than 100 mg of sediment per liter of water.

Zooming in on a roughly 10-square-kilometer region around Kinston, NC, she produced "false color" maps of the areas of open flooded areas and flooded areas covered with vegetation along the Neuse River. She also created maps to show relatively high and low sediment concentrations. On the map, zero indicates no sediment in the water, 100 milligrams per liter is an intermediate value, and 300 milligrams per liter is considered a high concentration.

As observed in the September 30, 1999, Landsat 7 image, surface sediment concentrations typically were in the range between 0 and 100 milligrams per liter in the Roanoke, Tar, and Neuse River channels and flooded areas. Although not directly comparable, depth-integrated concentrations of sediment reported by NAWQUA (United States Geological Survey) for the years 1993-1995 at the Kinston gauge on the Neuse River indicate a range of concentrations in the water column from less than 10 to a maximum of 79 milligrams per liter. Surface sediment concentrations are typically at least 2 to 3 times lower than average concentrations.

Neuse River in North Carolina
  Flooding along the Tar river

This pair of Landsat 7 images shows the before and after effects of Hurricane Floyd on the area where the Tar and Pamlico Rivers meet in Washington, NC. The top image was taken July 6, 1999, (before the hurricane) and the bottom was taken September 23, 1999.click to enlarge

Flood from different sensors

Therefore, says Mertes, although the concentrations in most of the rivers in this image were carrying relatively high sediment loads, the concentrations were not unusual. Given that the flooding experienced on September 30 followed three major storms in rapid sequence, she suspects sediment loads were lower on that date because much of the sediment available for erosion in the watersheds had already been transferred out of the system during the earlier storms.

"A lot of these floodwaters are probably left over from saturated flood regions created by Floyd that got filled up again," notes Mertes. "We're really looking at the legacy of Floyd. We see patches of clear water on flood plains and then some river water flowing into the flood plains.

"This case study is good for showing how antecedent flood waters affect inundation patterns," she continues, "and how flooding occurs over time. It demonstrates that floods often happen in ways we haven't thought of."

Although she can tell you how much sediment is suspended near the surface of floodwaters, she cannot say what the sediment is made of. Sediment "species analysis" is another technique she says other scientists can employ to identify types of sediment, but requires finer spectral resolution than is available with ETM+ data. She says MODIS data may provide sufficient spectral resolution to distinguish sediment type, and future study will test this hypothesis.

"At present, I don't do sediment species analysis," Mertes states, "but I would love for someone to take my images and look upstream to determine point sources enabling them to identify sediment types. That's one way these maps can be used."

References
Alsdorf, Douglas E., John M. Melack, Thomas Dunne, Leal A.K. Mertes, Laura L. Hess, & Laurence C. Smith, 2000: "Interferometric radar measurements of water level changes on the Amazon flood plain." Nature, Vol. 404, pp. 174-177.

Gomez, B., Leal A.K. Mertes, J.D. Phillips, F.J. Magilligan, and L.A. James, 1995: "Sediment Characteristics of an Extreme Flood - 1993 Upper Mississippi River Valley." Geology, Vol. 23, No. 11, pp 963-966.

Mertes, Leal A.K., 1997: "Documentation and significance of the perirheic zone on inundated floodplains." Water Resources Research, Vol. 33, pp. 1749-1762.

Mertes, Leal A.K, Milton O. Smith, and John B. Adams, 1993: "Estimating Suspended Sediment Concentrations in Surface Waters of the Amazon River Wetlands from Landsat Images." Remote Sens. Environ, Vol. 43, pp. 281-301.

Witte, W. G., C.H., W., Usry, J. W., Morris, W. D., and Gurganus, E. A., 1981, Laboratory measurements of physical, chemical and optical characteristics of Lake Chicot sediment waters.: NASA Technical Report, no. 1941, Hampton, Virginia, Langley Research Center.

back Assessing Floyd's Floods

 

These detailed maps of the Neuse River flooding in the Kinston, NC, area were produced using Landsat 7 data acquired on Sept. 30, 1999. In the top "false-color" image, red pixels indicate where there is turbid water, green shows vegetated or forested areas, yellow indicates pasture, blue shows clear water, and blue-green shows flooded forest lands.

In the middle image, the colors show more clearly the locations of open water (red pixels) and flooded vegetation (green). All other pixels are masked (black) for clarity. Comparing the middle image with the top image shows more clearly the areas of flooded vegetation that were difficult to discern in the top image and were therefore not classified as "flooded vegetation" due to limitations in the flood detection technique used.

The bottom image shows variations in the concentrations of suspended sediment near the surface of the Neuse River and the surrounding areas in flood. Red and yellow pixels show high sediment concentrations, greens are intermediate values, and blues indicate where the water contains low levels of sediment. Note that areas shown in this image are only open water areas; all other pixels are masked black for clarity. (Images courtesy Leal Mertes)


sediment concentration scale