Before the Flood

David Pacchioli
September 01, 1996

The river will have its day. The rains will come and the river will rise. The river will rise and spill its banks and spread, fanning hungrily over the lowest land, widening, seeking its rest. Spreading its chaos. An angry current, long pent up, unimaginable: a thundering coffee-brown tumble of mud-water, miles across, bobbing with tires and tree limbs, porch roofs and garbage cans. Mesmerizing. Running on, drowning cornfields, wheat fields, bean fields, baseball diamonds. Unrelenting. Sweeping through houses, picking up pick-up trucks, snatching up people and mailboxes, railroad ties and television antennas, chickens and dogs . . .

Then, suddenly, the anger will fade. The rains will stop. The sun will come out. For a while the water will remain—placid now, sorry even, plied by flat-bottomed rescue boats as it rolls back or soaks its way slowly into the ground. Leaving behind, eventually, everything it picked up along its route: swollen clapboards and stinking fish; barrels of oil and the best of the neighboring county's topsoil; raw sewage; tons and tons of pale river sand. In the state of Missouri, in the wake of the disastrous 1993 Mississippi River flood, hundreds of thousands of acres of farmland were left covered with up to eight feet of sand.

"That land will take hundreds of years to recover," says Ana Barros, Penn State assistant professor of civil and environmental engineering. "And that's a cost not included in the damage estimates."

Those estimates, by the U.S. Army Corps of Engineers, run to $10 billion. Sixteen thousand square miles of heartland—cities, towns, and farms—were submerged. Eight hundred levees were breached. Fifty people lost their lives.

"It is not a case of in two years we are all living happily again," Barros says. "The implications of a major flood are tremendous—they go way beyond what is covered by emergency funding. This is something that will be paid for by future generations."

The control of rivers has been a human preoccupation since earliest times. To harness that current to human purposes-- commercial navigation, irrigation—has been a central economic goal. Fertile bottomlands are crucial to food production. So people have engaged in taming the river: straightening, deepening, shortening, gathering, removing obstacles to passage. And turning back high water.

Seasonal flooding is part of a river's natural cycle. Keeping a river within its banks is an unnatural act, one that requires considerable will—and considerable infrastructure.

Beginning in the 1930s, the Mississippi River has undergone intensive and continuous improvement, with construction of levees and reservoirs and channels and locks. The Flood Control Act of 1937 gave the federal government a legal role in regional flood control and created the Soil Conservation Service to protect agricultural land.

In the frequently swampy lower Mississippi basin, a nearly continuous system of levees arose, set back thousands of feet from the river. The spacing provided plenty of room for flood waters to dissipate. The system has worked well.

"This is why one rarely hears of flooding along the lower Mississippi anymore," Barros says.

The upper Mississippi has been a different story. There, the levees were placed very close to the river so that the rich land that had once been part of the floodplain could be appropriated for agriculture. In places the river was deepened, to further increase its capacity. Elsewhere, channels were cut to straighten its meandering flow. Between 1930 and 1940, the river was shortened by 243 kilometers. The river was narrowed—by tight levees and wing dikes, which cut off side channels, in some places by 75 percent. Upstream, north of St. Louis, reservoirs were placed for storing excess water along the Iowa, Minnesota, and Kansas Rivers.

Against normal, seasonal flooding, these measures have worked pretty well. But they have also wrought major changes on the river, changes which upset the balance of the river system and, when it comes to larger events, have had increasingly negative impacts.

"The real problem in a flood," says Barros, is not the water itself. It's water flowing fast, carrying a lot of debris with it, and washing away fertile soil."

Heavy sediment deposits downstream limit the river's capacity for carrying high flows. The loss of side channels, closed off either purposely with dikes or incidentally with sediments, eliminates safety valves which could help drain off flood water.

The concommitant destruction of wetlands has drastically lowered floodplain holding capacity. Wetlands act as a sponge in times of flood, soaking up extra water and releasing it gradually as conditions permit. In Missouri, representative of the region, wetland acreage over the last 200 years has shrunk from almost 5 million acres to less than 700,000.

The upper Mississippi, Barros notes, is now fundamentally a channel, not a river. A channel, she notes, has no capacity to adapt to variable conditions. Tamed, constricted, "It can't evolve to prepare itself for the next event. This river has nowhere to go."

When a big flood comes along, this inflexibility translates into more damage to the surrounding landscape. Barros cites a comparison of two similar floods along the Mississippi, one occurring in 1908 and the other during 1973, which showed that despite similar amounts of rainfall and nearly identical conditions, the 1973 flood resulted in downstream water levels that were eight feet higher.

In April of 93, the spring rains came and kept coming. Four months worth of solid rain brought record levels of precipitation over Illinois, Iowa, Kansas, Minnesota, Missouri, Nebraska, North Dakota, South Dakota, and Wisconsin. The sheer volume of water was trouble enough; what made things worse, Barros says, is that the river had no way to handle it all. Confined, moving at high velocity, the sediment-heavy flow rose and rose. Upstream, to boot, a loose network of reservoirs effectively backfired: filled to the brim with record rains, the reservoirs had to be released all at once, to prevent their bursting.

The result downstream was a foregone conclusion. Flood gauges in St.Louis topped out at 49 feet—a full 20 feet above the previous record. Eight hundred out of 1300 levees were overtopped, causing flooding much worse, Barros suggests, than if there had been no levees at all.

The much-abused upper Mississippi, in short, forcefully reclaimed its floodplain.

A large-scale flood is a complex event, the result of a rare confluence of dynamic factors.

"Floods are largely a surface hydrological process," Barros says, "but they can't be separated from atmospheric phenomena. They involve spatial and temporal scales of weather systems as well as spatial and temporal variations of land surface."

Yet, Barros notes, "Most flood control design is reactive—done in response to specific events. It lacks adequate consideration of long-term, system-wide effects."

Take climatic variability. Global climate is determined to some extent by the El Nino, the occasional appearance of unusually warm ocean temperatures along the west coast of South America. The El Nino varies in intensity on a scale of every few years.

During the period between 1947 and 1968, Barros notes, there were a number of strong El Nino events, whose effect was to increase flooding in the upper Mississippi Valley. The river's response was to widen in places, accommodating the extra water. Wetlands spread out across the floodplain.

The human response to this temporary adaptation, she found, did not take climate cycles into account. (For one thing, not much was known about El Nino's effects back then.) On one section of the river that Barros studied, engineers built a large wing dike. The result, when a drier period resumed, was a drastic narrowing of channel width—three-quarters of a mile over 15 years.

Accurate forecasting of large-scale floods would require not only precise climatic modeling (and we know how difficult it is to predict the weather), and precise hydrological modeling, but a precise meshing of the two. "This boundary area, hydrometeorology, has been overlooked for a long time," Barros says.

Existing flood forecasts, Barros says, are largely based on historical stream flow data and precipitation averages. "Hydrological models are very simplified, for the purpose of quick forecasting.

"Because hydrologic phenomena are so complex and we dont know how to describe this complexity," Barros says, "we rely on the safety net of randomness. We treat things as if there were a high degree of randomness and a low degree of complexity, when in reality there is a high degree of complexity and a low degree of randomness. That's the choice we have made."

The other safety net for current hydrological models, she says, is their stationary quality. "We assume that the processes involved are unchanging—that things will happen in the future as they have in the past. We know this not to be true, but we proceed as if it were."

Thus, Barros says, available flood forecasting is very short term: on the order of a few hours. Long-term worst-case estimates, on the basis of which engineering designs are made, are founded on the same scant data. Not surprisingly, those estimates are often exceeded in the event.

The '93 flood was by engineering standards a freak, a 500-year flood by some estimates. That does not mean, Barros says, that it won't happen again next year. It only means there's a 1 in 500 chance.

Barros is working on more realistic flood forecasts. It hasn't been easy. Floods being judged an occasional menace, no one wants to spend the money for the necessary research. But now we have the tools, she says, to make real changes.

"We have the technology to look at the data as we could not before. Advances in data analysis allow us to see in four dimensions—looking at space and time at once, rather than tracking one point over time."

With such capabilities, Barros and her colleagues can more accurately depict the many factors involved in a major flood. Factors like spatial distribution of precipitation.

Even a major flood can be a matter of very localized precipitation: exceedingly heavy rain in a few key watersheds can be a deciding factor. Point gauges and radar are not enough to measure precipitation acccurately at this scale. Mountainous terrain, for example, can easily defeat radar. At the same time, even low mountains like Pennsylvania's can have a profound effect on local rainfall.

Barros won a national award for her 1993 Ph.D. dissertation on these so-called orographic effects. "The scales of orographic precipitation are pretty small. One hundred meters can be the difference between heavy snow and rainfall. To understand it, you need to work at very fine resolution. We have developed an algorithm which preserves resolution even at larger spatial scales, i.e., continents.

"I believe orographic effects also affect the Missouri," she adds. She looked at about 300 basins in the Missouri valley, measuring to see whether the '93 flood was the biggest ever. "In most of the large basins," she says, "the answer is no. In some where orographic effects are important, however, the answer was yes."

Barros has demonstrated convincingly that orographic rainfall played an important role in the Susquehanna River flood that inundated large sections of Pennsylvania on January 19, 1996, killing fourteen people and causing $1 billion in damages.

In the narrow Appalachian valleys above Harrisburg, between Williamsport and Wilkes-Barre, air masses were forced to ascend by the surrounding steep walls, causing condensation, and rainfall intensities were much higher than elsewhere. Because those basins are so small, almost all the water was immediately converted to run-off.

Barros and her graduate students looked at the Loyalsock basin, near Lock Haven, as a case in point. "There is no precipitation gauge in the basin," Barros says, "and early on the radar showed nothing, only some low convective cells." Meanwhile, however, actual precipitation amounts at the ground were very high. "And by the time the front finally came through, the snowpack was almost all gone—so all the rainfall was converted to runoff."

Along with the high water came one of those confounding factors, especially rare for a flood in Pennsylania: ice dams.

"What happens," Barros says, "is the ice piles up and restricts flow to a narrow channel below the ice until the pressure builds up high enough to burst it." Below such a stricture, as it builds, the paucity of flow confounds stream gauges. At Loyalsock, one large ice dam held back 13 feet of water before it burst, causing water levels downstream at Harrisburg to rise nine feet in four hours—most of that in a single hour. "Harrisburg typically has eight hours of lead time before a flood hits," Barros says. "In this case, it had three.

"We need to be able forecast when these dams will form, and how they will grow. This is probably a random process. It's very hard to look at—it defies traditional knowledge. But it's a critical problem."

Barros and her students are also looking at larger scale effects.

"We're trying to come up with indexes that would allow us to forecast a number of conditions that would lead to the occurrence of very large floods over continental areas, on a timescale of one to two years. We want to develop methods to use information from general circulation models (GCMs) to infer what would actually happen in a river basin."

GCMs make predictions on a global basis. Over smaller areas they're not very useful. "The hydrology in them is very crude," Barros says. "Their precipitation predictions are purely averaged—they don't match any ground observations anywhere. But we depend on these models to learn about climate change, and how we should be planning land use for the coming years.

"To bring GCMs up to speed in this area, you'd have to include the oceans," she continues. "Doing so in a reliable sense will require much more computer power than we currently have.

"We are looking at another approach. Instead of looking at precipitation values at given points, we are looking at how spatial structures evolve over time."

To wit: Cloud cover differs substantially in wet and dry years. Clouds affect the radiation budget at the land surface, decreasing the amount of solar radiation coming in, and trapping what's going out. In a cloudier year, evapotranspiration—the amount of moisture returned to the atmosphere by evaporation—goes down. Soil moisture is retained in the soil. The land surface is wetter.

"It's like a bucket. If you stop evapotranspiration and precipitation keeps on, the bucket fills up, and eventually runs over. Looked at over extended areas, and not through point precipitation, these relationships should make sense.

"On a global scale, El Nino has a strong impact on cloudiness over the continents—its extent, direction, spatial arrangement. You have to factor this in with precipitation data if you want to understand flooding."

High precipitation, Barros says, doesn't necessarily mean a superflood. "Over large basins like the Susquehanna, the Ohio, the Mississippi, the Missouri, you need special conditions—a whole system.

"Big floods usually happen in cooler years."

In standard engineering practice, when a flood-control structure—a dam or a levee—is built, its designers first estimate, for a given area, the worst possible storm that could occur, and from that the worst possible flood. To do so, they focus mainly on river hydraulics and average streamflow data. How often have there been floods in the past, and how severe?

This approach can't account for the effects of changes in climate, or in land-use. To do so, you need to model the interactions between hydrologic, geomorphic, and atmospheric processes.

Barros, with support from the National Science Foundation, is working on such an approach, an integrated computer network which will couple hydrometeorology, soil erosion, and river flow models.

"With the tools we have now," she says, "we can look at different scenarios. We can do what-if experiments. We can take into account variability."

Integrated models should provide longer headstarts for dealing with "flash" floods like the January episode on the Susquehanna. They should also improve long-term forecasting, providing better design criteria for flood control.

But no reasonable amount of structural engineering would have stopped the Mississippi deluge.

The point is important, Barros notes, because the big flood raised big questions for those concerned with protecting the upper Mississippi basin. Should they go back in and rebuild as before—only bigger this time—or should they instead step back and take stock?

Against the $10 billion damage figure for the '93 flood, the Army Corps of Engineers claims that existing flood controls saved the basin damages of $20 billion more. "But this figure assumes the same land use would have been in place, so it's not realistic. What we really need is to do a realistic cost-benefit analysis."

For Barros, part of understanding the river is learning to respect it: recognizing that ultimately it will not be controlled. "We must learn to work with the river instead of against it." This means recognizing the river as a complex, self-regulating system, and seeking to restore as much of its integrity as possible. At the same time, she says, "We have to anticipate the worst, and design systems that work well in failure."

A "non-structural" approach to flood control engineering on the upper Mississippi, Barros suggests, could include a system of secondary levees, set well back from the river, "like the natural levees on the lower Mississippi." A system of channels could be dug to guide overflow to where it does the least harm, spreading it out and slowing it down. Catch basins could be built to replace some of the absorptive capacity of lost wetlands. And planting trees along the river would help to impede erosion.

Such measures, she acknowledges, would require substantial sacrifice. "They would take away some of the most productive land, and make the rest of the floodplain more difficult to work."

Admittedly, too, this kind of approach goes against the conventional engineering grain. "When I was in school," Barros admits, "if you weren't working on some big hydraulics project, building a big dam or a big bridge, you were nothing. The idea was to go to the landscape and put a big fingerprint on it.

"We can't afford that approach any more."

In pre-modern days, Barros notes, in Egypt, Mesopotamia, China, "They knew they didn't have the technology to conquer, so they adapted—by listening to what the river and the landscape were telling them." In China, at the Dujiangyan waterworks on the Changjiang River, an ancient design is still successfully being used. "Here, most of the time, the river flows where humans want it to. On the rare occasions when it floods, it goes where it wants—but the design allows for that to occur without too much destruction."

For a river like the 21st-century Mississippi, with so much invested in the modern way of doing things, a back-to-the-future restoration would admittedly require a tremendous effort. There's a limit, Barros acknowledges, to what can be undone.

On the other hand, she writes, "Most of the viable structural engineering of large rivers has been completed." Or to put it another way: There is not a lot left to do to the river in conventional terms to be ready for next time.

And if there is one thing Barros is certain of, it is that there will be a next time.

Ana Paula Barros, Ph.D., is assistant professor of civil and environmental engineering in the College of Engineering, 213C Sackett Building, University Park, PA 16801; 814-863-8609. Barros received a National Science Foundation Faculty Early Career Development Program Award in 1996 for her research proposal, "Integrated Hydrologic Analysis for Flood Forecasting and Control."

Last Updated September 01, 1996